def run(self): while not self.done: logging.info("Scanning for new devices...") dev_list = UvcCapture.device_list() # Find available devices if self.dev_list != dev_list: # If found a new device or an old device was disconnected self.dev_list = dev_list # Update the list of devices self.sig_new_dev.emit() # And fire a signal print "Compatible devices found:\n\t" + str(self.dev_list) self.sleep(self.REFRESH_RATE)
def refresh_cam_device_list(): old_dev_selected = window.drpInput.currentText() window.drpInput.clear() window.drpInput.addItem("None") # Add a 'None' item so user can turn off camera for dev in threadDevLister.dev_list: txt = UvcCapture.get_human_friendly_identifier(dev['name'], dev['idVendor'], dev['idProduct'], dev['serialNumber']) window.drpInput.addItem(txt) if txt == old_dev_selected: window.drpInput.setCurrentIndex(window.drpInput.count()-1) threadFrameGrabber.sig_new_dev_selected.emit(window.drpInput.currentText())
def producer(v, v2, q, arr): cap = UvcCapture(0) cap.select_best_frame_mode(60) cap.load_settings("UVCcam settings - USB 2.0 Camera.txt") q.get() # Wait for opencv to init t = [] last_t = datetime.now() print "OpenCV init'ed! (consumer process says so)" while v.value < 300: t.append(datetime.now()) frame = cap.get_frame_robust() t.append(datetime.now()) img = frame.bgr t.append(datetime.now()) arr[:] = frame.bgr v2.value = frame.index t.append(datetime.now()) ### q.put([img, frame.index]) # q.put(frame.index) t.append(datetime.now()) print "\t\t\tPRODUCER: At t={} picture #{:04d} was taken; totalDeltaT={}ms;\t{}".format(str(t[3].time())[:-3], frame.index, (t[-1]-last_t).total_seconds()*1000, "\t\t".join(["{}->{}: {}ms".format(i+1, i+2, (t[i+1]-t[i]).total_seconds()*1000) for i in range(len(t)-1)])) last_t = t[-1] t = [] q.put("STOP") print "Producer exiting with value {}".format(v.value)
def init_video_cam_and_cv_algorithm(self, create_video_folder=True): """ Initializes camera: connects to it, loads settings from config file (if available), loads color threshold and blob detector settings (if available), creates a folder to save cv output images... :param create_video_folder: True to create folder specified by VIDEO_FOLDER (where output frames will be saved) """ self.video_capture = UvcCapture.new_from_settings(self.CAMERA_SETTINGS_FILE) # Connect to device specified by settings, and load its desired param values if self.video_capture is None: # If unable to connect to device specified by settings, open first available camera self.video_capture = UvcCapture(0) if self.video_capture is None: # If still unable to connect, raise an exception raise Exception("Couldn't open camera! :(") # If we're here, we couldn't connect to device specified by settings but were able to open 1st available cam if not self.video_capture.load_settings(self.CAMERA_SETTINGS_FILE): # Try to load frame size & rate from settings self.video_capture.select_best_frame_mode(60) # If loading settings failed, choose best frame size with fps >= 60 logging.info("Camera opened! :)") # Initialize cv algorithm too: load settings for color thresholding and blob detector self.load_color_thresh_settings() self.load_blob_detector_settings() # Sometimes, first couple frames take a long time to be obtained, do it before quad goes in flying mode self.video_capture.get_frame_robust() self.video_capture.get_frame_robust() # Initialize PID setpoints and initial input value to the center of the frame self.cf_PID_roll.setSetPoint(self.video_capture.frame_size[0] / 2) self.cf_PID_thrust.setSetPoint(self.video_capture.frame_size[1] / 2) self.cf_PID_pitch.setSetPoint(40) self.cf_PID_roll.PIDpos.curr_input = self.cf_PID_roll.getSetPoint() self.cf_PID_thrust.PIDpos.curr_input = self.cf_PID_thrust.getSetPoint() self.cf_PID_pitch.PIDpos.curr_input = self.cf_PID_pitch.getSetPoint() # Prepare the folder self.VIDEO_FOLDER so we can store each frame we processed (for debugging) if create_video_folder: shutil.rmtree(self.VIDEO_FOLDER, ignore_errors=True) # Delete the folder and its contents, if it exists (ignore errors if it doesn't) os.makedirs(self.VIDEO_FOLDER) # Now create the folder, which won't throw any exceptions as we made sure it didn't already exist
class CamFrameGrabberThread(QThread): """ This class implements a background task that constantly grabs new frames from the selected camera. """ DEFAULT_CAM_NAME = "None" # "USB 2.0 Camera" DEFAULT_CAM_VEND_ID = -1 # 1443 DEFAULT_CAM_PROD_ID = -1 # 37424 DEFAULT_CAM_SERIAL_NUM = None DEFAULT_FPS = 60 FPS_ESTIMATION_ALPHA = 0.98 FRAME_SIZE_STR_SPLITTER = " x " sig_update_cam_ui = pyqtSignal() # Signal that's emitted whenever the UI associated to the camera controls needs to be refreshed (eg: new frame size selected -> need to resize window, error grabbing camera frame -> disconnect camera...) sig_new_dev_selected = pyqtSignal(str) # Signal that's emitted when a device is selected (could be the device that was already selected). It's emitted either from the UI when the user changes the selected item in drpInput, or from the code when sig_new_dev is emitted (as a new device is connected or an old one disconnected) sig_new_img = pyqtSignal() # Signal that's emitted every time a new camera frame becomes available sig_error = pyqtSignal(str) # Signal that's emitted every time an error occurred def __init__(self): QThread.__init__(self) self.done = False self.dev_selected_name = self.DEFAULT_CAM_NAME self.dev_selected_vend_id = self.DEFAULT_CAM_VEND_ID self.dev_selected_prod_id = self.DEFAULT_CAM_PROD_ID self.dev_selected_serial_num = self.DEFAULT_CAM_SERIAL_NUM self.cap = None self.qPix = QPixmap(1, 1) self.cv = Spotter() self.cv.workers.append(FakeWorkerDrone()) # Add a fake worker (we just need the cv algorithm to look for 1 ping pong ball) self.cv.video_capture = FakeVideoCapture() self.cv.world_to_camera_transf = np.hstack((np.eye(3), np.zeros((3, 1)))) self.cv.load_color_thresh_settings() self.cv.load_blob_detector_settings() self.cv_do_color = True self.cv_do_blob = True self.cv_evaluate_px_at = np.array([0, 0]) self.cv_evaluate_px_value = np.array([0, 0, 0], dtype=np.uint8) self.index = 0 self.actual_fps = 0 def __del__(self): self.wait() @pyqtSlot() def stop_thread(self): # This method is used to externally request this thread to terminate self.done = True # Flag to indicate the outer while loop in run_experiment() to finish thread execution self.change_selected_device("EXIT! :P") # Choose a void device, to force the inner while loop in run_experiment() to exit after grabbing at most one frame if self.wait(5000) is not True or self.cap is not None: # Worst-case scenario: force quit if self.cap is still not None after 5sec logging.warning("Uh-ooohh, CamFrameGrabberThread didn't terminate even after waiting for 5sec. Force quitting :S") self.cap = None self.terminate() else: logging.info("CamFrameGrabberThread exited cleanly :)") @pyqtSlot(str) def change_selected_device(self, dev_selected): if self.dev_selected_name != dev_selected: logging.info("New device selected: {}".format(dev_selected)) self.dev_selected_name, self.dev_selected_vend_id, self.dev_selected_prod_id, self.dev_selected_serial_num = UvcCapture.parse_human_friendly_identifier(dev_selected) @pyqtSlot(str, bool) def change_cam_frame_size(self, new_frame_size, emit_signal): if self.cap is not None and window.drpSize.count() > 0: # Sanity check (in case capture device was just closed or we're clearing drpSize to re-add available sizes) logging.info("Changing frame size to {}".format(new_frame_size)) self.cap.do_undistort = False # Pause undistortion while frame_size is changing to avoid problems in the frame capture thread self.cap.frame_size = tuple([int(x) for x in new_frame_size.split(self.FRAME_SIZE_STR_SPLITTER)]) self.cap.init_undistort_maps(self.cap.calib_file) if emit_signal: # If the user was the one who clicked on this combobox item (as opposed to me, manually from the code), update cam UI self.sig_update_cam_ui.emit() # drpFPS might need to be updated (different frame sizes might have different fps available) @pyqtSlot(str) def change_cam_frame_rate(self, new_frame_rate): if self.cap is not None and window.drpFPS.count() > 0: # Sanity check (in case capture device was just closed or we're clearing drpSize to re-add available sizes) logging.info("Changing frame rate to {}".format(new_frame_rate)) self.cap.frame_rate = int("".join(x for x in new_frame_rate if x.isdigit())) # Keep only the numbers (= remove the " fps" part, if exists) @pyqtSlot(str, int, bool) def change_cam_control_setting(self, ctrl_name, new_value, is_bool): """ Process a request (from the UI) to change a particular camera setting. :param ctrl_name: Particular camera setting that wants to be modified (eg: exposure time, white balance...) :param new_value: New value desired for the camera setting :param is_bool: True if the UI control is a CheckBox, False if it's a SpinBox (ie: TextBox with up&down arrows) """ if self.cap is not None: # Sanity check (in case capture device was just closed) for c in self.cap.controls: if c.display_name == ctrl_name: try: if is_bool: c.set_value(c.min_val if (new_value == Qt.Unchecked) else c.max_val) else: c.set_value(new_value) logging.info("'{}' changed to {}".format(c.display_name, c.value)) except: self.sig_error.emit("Unable to change '{}' property to '{}'! Make sure the value is valid (within bounds, not disabled by other setting like 'Auto' modes, etc).".format(ctrl_name, new_value)) return @pyqtSlot(bool) def set_cv_do_color(self, checked): self.cv_do_color = checked window.grpBlob.setEnabled(checked) self.set_cv_do_blob(False if not checked else window.grpBlob.isChecked()) @pyqtSlot(bool) def set_cv_do_blob(self, checked): self.cv_do_blob = checked @pyqtSlot(int, str, bool) def change_HSV_setting(self, new_value, letter, is_max): """ Updates the corresponding field (H,S or V variable, min or max) of the HSV color thresholding filter. :param new_value: New value to set the corresponding threshold to :param letter: either 'H', 'S', or 'V' (case insensitive), depending on which threshold should be modified :param is_max: True to change the upper threshold; False to change the lower one """ letter = letter.lower() # To avoid problems, convert the letter to lowercase and only compare to lowercase index = 0 if letter == 'h' else 1 if letter == 's' else 2 # Convert H, S, V to 0, 1, 2 # Now update the right variable based on is_max if is_max: self.cv.cv_HSV_thresh_max[index] = new_value else: self.cv.cv_HSV_thresh_min[index] = new_value @pyqtSlot(int, str, bool) def change_blob_setting(self, new_value, letter, is_max): """ Updates the corresponding setting (area, circularity, convexity or inertia, min or max) of the blob detector. :param new_value: If int: New value to set the corresponding threshold to; if bool: activate/deactivate filter :param letter: either 'A', 'C', 'V', or 'I' (case insensitive), depending on which setting should be modified :param is_max: True to change the upper threshold; False to change the lower one [ignored if new_value is bool] """ letter = letter.lower() # To avoid problems, convert the letter to lowercase and only compare to lowercase param = "Area" if letter == 'a' else "Circularity" if letter == 'c' else "Convexity" if letter == 'v' else "Inertia" if type(new_value) is bool: setattr(self.cv.cv_blob_detect_params, "filterBy{}".format(param), new_value) else: setattr(self.cv.cv_blob_detect_params, "{}{}{}".format("max" if is_max else "min", param, "Ratio" if letter == 'i' else ""), new_value) def run(self): while not self.done: self.sleep(1) # Outer while loop just waits until a device is selected self.cap = UvcCapture(self.dev_selected_name, self.dev_selected_vend_id, self.dev_selected_prod_id, self.dev_selected_serial_num) if self.cap is None: # If we didn't find the desired cam, don't continue self.qPix = QPixmap(1, 1) # Default to a black pixel self.cv_evaluate_px_value = np.array([0, 0, 0], dtype=np.uint8) logging.info("No compatible cameras found or chosen camera name not available :(") self.sig_update_cam_ui.emit() # Make sure camera-related UI is updated (eg: don't show exposure time control if no camera is selected...) continue # If we made it here, we successfully connected to the selected camera. Print camera info and select default values for c in self.cap.controls: logging.info("\t{} = {} {} (def:{}, min:{}, max:{})".format(c.display_name, str(c.value), str(c.unit), str(c.def_val), str(c.min_val), str(c.max_val))) self.cap.select_best_frame_mode(self.DEFAULT_FPS) logging.info(self.cap.name + " has the following available modes:\n\t" + str([tuple(x) for x in self.cap.sorted_available_modes()]) + "\nSelected mode: " + str(self.cap.frame_mode)) # self.cap.print_info() logging.debug("LOADING SETTINGS returned {}".format(self.cap.load_settings(self.cv.CAMERA_SETTINGS_FILE))) self.sig_update_cam_ui.emit() # Update camera-related UI (eg: add controls for exposure time, white balance, etc.) img = np.zeros(self.cap.frame_size) # Resize qPix to fit new image size self.qPix = QPixmap.fromImage(QImage(img.data, img.shape[1], img.shape[0], QImage.Format_RGB888)) self.actual_fps = self.cap.frame_rate tt = datetime.now() # Initialize tt (used to estimate actual frame rate) to prevent an error on the first loop iteration while self.cap.name == self.dev_selected_name and self.cap.vend_id == self.dev_selected_vend_id and self.cap.prod_id == self.dev_selected_prod_id and self.cap.serial_num == self.dev_selected_serial_num: # Run an inner while loop to capture frames as long as the selected device is kept constant ok = False t = datetime.now() for a in range(3): try: frame = self.cap.get_frame_robust() self.index = frame.index ok = True break except Exception as e: logging.error('DEBUG - Could not get Frame. Error: "{}". Tried {} time{}.'.format(e.message, a+1, "s" if a > 0 else "")) if not ok: self.sig_error.emit("Couldn't get camera frame after 3 tries, reconnecting...") break # Exit inner loop (force an iteration on the outer loop) after 3 consecutive failed attempts to grab a frame # We successfully grabbed a frame, let's store the value of the pixel cv_evaluate_px_at for debugging purposes try: self.cv_evaluate_px_value = np.array(frame.bgr[self.cv_evaluate_px_at[1], self.cv_evaluate_px_at[0]]) # Create a new np.array so that the value at the pixel is not cached, but copied (ie: if frame.bgr is modified, cv_evaluate_px_value is not). Also note that images should be indexed [y,x]! except: self.cv_evaluate_px_at = np.array([0, 0]) # Now we can run the CF detection algorithm on it (if grpColor is checked) and modify frame.bgr if we want t2 = datetime.now() if self.cv_do_color: cf_curr_pos = self.cv.detect_cf_in_camera(frame.bgr, self.cv_do_blob)[0] # Run CF detection (and blob detection if checked) # Now overlay the mask over the camera frame t3 = datetime.now() if cf_curr_pos is not None and self.cv_do_blob: cv2.circle(self.cv.cv_cam_frame, tuple(cf_curr_pos[0:2].astype(int)), int(cf_curr_pos[2]+5), [0, 255, 0], -1) # np.putmask(self.cv.cv_cam_frame[:,:,0], self.cv.cv_filtered_HSV_mask, 255) # np.putmask(self.cv.cv_cam_frame[:,:,1], self.cv.cv_filtered_HSV_mask, 0) # np.putmask(self.cv.cv_cam_frame[:,:,2], self.cv.cv_filtered_HSV_mask, 255) mask = cv2.cvtColor(self.cv.cv_filtered_HSV_mask, cv2.COLOR_GRAY2BGR).astype(bool) np.putmask(self.cv.cv_cam_frame, mask, np.array([255, 0, 255], dtype=np.uint8)) else: t3 = datetime.now() # For consistency, save current time as t3 (even if we didn't do anything # And finally convert the resulting image to qPix and emit the appropriate signal so the UI can refresh it t4 = datetime.now() img = cv2.cvtColor(frame.bgr, cv2.COLOR_BGR2RGB) self.qPix.convertFromImage(QImage(img.data, img.shape[1], img.shape[0], QImage.Format_RGB888)) self.sig_new_img.emit() t5 = datetime.now() self.actual_fps = self.FPS_ESTIMATION_ALPHA*self.actual_fps + (1-self.FPS_ESTIMATION_ALPHA)/(t5-tt).total_seconds() logging.debug("At t={} picture #{:03d} was taken; deltaTtotal={:6.2f}ms [{:6.2f}ms] (capture_frame->{:6.2f}ms + detect->{:6.2f}ms + blob->{:6.2f}ms + putmask->{:6.2f}ms + toQPixmap->{:6.2f}ms) -> Estimated FPS: {:.2f}".format(t2, frame.index, (t5-tt).total_seconds()*1000, (t5-t).total_seconds()*1000, (t2-t).total_seconds()*1000, (self.cv.t_events[-2]-t2).total_seconds()*1000 if self.cv_do_color else 0, (self.cv.t_events[-1]-self.cv.t_events[-2]).total_seconds()*1000 if self.cv_do_blob else 0, (t4-t3).total_seconds()*1000, (t5-t4).total_seconds()*1000, self.actual_fps)) tt = datetime.now() self.cap = None # Set cap to None to automatically release & disconnect from the camera
def run(self): while not self.done: self.sleep(1) # Outer while loop just waits until a device is selected self.cap = UvcCapture(self.dev_selected_name, self.dev_selected_vend_id, self.dev_selected_prod_id, self.dev_selected_serial_num) if self.cap is None: # If we didn't find the desired cam, don't continue self.qPix = QPixmap(1, 1) # Default to a black pixel self.cv_evaluate_px_value = np.array([0, 0, 0], dtype=np.uint8) logging.info("No compatible cameras found or chosen camera name not available :(") self.sig_update_cam_ui.emit() # Make sure camera-related UI is updated (eg: don't show exposure time control if no camera is selected...) continue # If we made it here, we successfully connected to the selected camera. Print camera info and select default values for c in self.cap.controls: logging.info("\t{} = {} {} (def:{}, min:{}, max:{})".format(c.display_name, str(c.value), str(c.unit), str(c.def_val), str(c.min_val), str(c.max_val))) self.cap.select_best_frame_mode(self.DEFAULT_FPS) logging.info(self.cap.name + " has the following available modes:\n\t" + str([tuple(x) for x in self.cap.sorted_available_modes()]) + "\nSelected mode: " + str(self.cap.frame_mode)) # self.cap.print_info() logging.debug("LOADING SETTINGS returned {}".format(self.cap.load_settings(self.cv.CAMERA_SETTINGS_FILE))) self.sig_update_cam_ui.emit() # Update camera-related UI (eg: add controls for exposure time, white balance, etc.) img = np.zeros(self.cap.frame_size) # Resize qPix to fit new image size self.qPix = QPixmap.fromImage(QImage(img.data, img.shape[1], img.shape[0], QImage.Format_RGB888)) self.actual_fps = self.cap.frame_rate tt = datetime.now() # Initialize tt (used to estimate actual frame rate) to prevent an error on the first loop iteration while self.cap.name == self.dev_selected_name and self.cap.vend_id == self.dev_selected_vend_id and self.cap.prod_id == self.dev_selected_prod_id and self.cap.serial_num == self.dev_selected_serial_num: # Run an inner while loop to capture frames as long as the selected device is kept constant ok = False t = datetime.now() for a in range(3): try: frame = self.cap.get_frame_robust() self.index = frame.index ok = True break except Exception as e: logging.error('DEBUG - Could not get Frame. Error: "{}". Tried {} time{}.'.format(e.message, a+1, "s" if a > 0 else "")) if not ok: self.sig_error.emit("Couldn't get camera frame after 3 tries, reconnecting...") break # Exit inner loop (force an iteration on the outer loop) after 3 consecutive failed attempts to grab a frame # We successfully grabbed a frame, let's store the value of the pixel cv_evaluate_px_at for debugging purposes try: self.cv_evaluate_px_value = np.array(frame.bgr[self.cv_evaluate_px_at[1], self.cv_evaluate_px_at[0]]) # Create a new np.array so that the value at the pixel is not cached, but copied (ie: if frame.bgr is modified, cv_evaluate_px_value is not). Also note that images should be indexed [y,x]! except: self.cv_evaluate_px_at = np.array([0, 0]) # Now we can run the CF detection algorithm on it (if grpColor is checked) and modify frame.bgr if we want t2 = datetime.now() if self.cv_do_color: cf_curr_pos = self.cv.detect_cf_in_camera(frame.bgr, self.cv_do_blob)[0] # Run CF detection (and blob detection if checked) # Now overlay the mask over the camera frame t3 = datetime.now() if cf_curr_pos is not None and self.cv_do_blob: cv2.circle(self.cv.cv_cam_frame, tuple(cf_curr_pos[0:2].astype(int)), int(cf_curr_pos[2]+5), [0, 255, 0], -1) # np.putmask(self.cv.cv_cam_frame[:,:,0], self.cv.cv_filtered_HSV_mask, 255) # np.putmask(self.cv.cv_cam_frame[:,:,1], self.cv.cv_filtered_HSV_mask, 0) # np.putmask(self.cv.cv_cam_frame[:,:,2], self.cv.cv_filtered_HSV_mask, 255) mask = cv2.cvtColor(self.cv.cv_filtered_HSV_mask, cv2.COLOR_GRAY2BGR).astype(bool) np.putmask(self.cv.cv_cam_frame, mask, np.array([255, 0, 255], dtype=np.uint8)) else: t3 = datetime.now() # For consistency, save current time as t3 (even if we didn't do anything # And finally convert the resulting image to qPix and emit the appropriate signal so the UI can refresh it t4 = datetime.now() img = cv2.cvtColor(frame.bgr, cv2.COLOR_BGR2RGB) self.qPix.convertFromImage(QImage(img.data, img.shape[1], img.shape[0], QImage.Format_RGB888)) self.sig_new_img.emit() t5 = datetime.now() self.actual_fps = self.FPS_ESTIMATION_ALPHA*self.actual_fps + (1-self.FPS_ESTIMATION_ALPHA)/(t5-tt).total_seconds() logging.debug("At t={} picture #{:03d} was taken; deltaTtotal={:6.2f}ms [{:6.2f}ms] (capture_frame->{:6.2f}ms + detect->{:6.2f}ms + blob->{:6.2f}ms + putmask->{:6.2f}ms + toQPixmap->{:6.2f}ms) -> Estimated FPS: {:.2f}".format(t2, frame.index, (t5-tt).total_seconds()*1000, (t5-t).total_seconds()*1000, (t2-t).total_seconds()*1000, (self.cv.t_events[-2]-t2).total_seconds()*1000 if self.cv_do_color else 0, (self.cv.t_events[-1]-self.cv.t_events[-2]).total_seconds()*1000 if self.cv_do_blob else 0, (t4-t3).total_seconds()*1000, (t5-t4).total_seconds()*1000, self.actual_fps)) tt = datetime.now() self.cap = None # Set cap to None to automatically release & disconnect from the camera
def change_selected_device(self, dev_selected): if self.dev_selected_name != dev_selected: logging.info("New device selected: {}".format(dev_selected)) self.dev_selected_name, self.dev_selected_vend_id, self.dev_selected_prod_id, self.dev_selected_serial_num = UvcCapture.parse_human_friendly_identifier(dev_selected)
def estimate_cam_to_world_transform(calibration_file, is_chessboard=False, cell_size=0.02): """ Opens a GUI and displays the camera's output. For every frame, it also looks for a calibration pattern (eg: OpenCV's chessboard pattern), and computes the cam->world transformation matrix if found. :param calibration_file: Str indicating the name of the calibration file to load camera_matrix and dist_coefs from :param is_chessboard: True if the calibration pattern is a chessboard. False, for asymmetric circle grid :param cell_size: Float indicating the distance (in m) between two consecutive points in the pattern grid """ cam = UvcCapture.new_from_settings(Spotter.CAMERA_SETTINGS_FILE) if cam is None: logging.error( "Please connect the camera, can't do this without you :P") return cam.do_undistort = False # Don't undistort, we're already taking into account camera_matrix in the equations # for c in cam.controls: # if c.display_name == "Sharpness": # c.value = c.max_val # break win_title = "Estimating camera->world coordinate transform" cv2.namedWindow(win_title) cv2.moveWindow(win_title, 100, 100) while cv2.waitKey(1) < 0: t_start = datetime.now() img = cam.get_frame_robust().bgr world_to_camera_transf, F = auxV.find_world_to_cam_and_F( img, calibration_file, is_chessboard=is_chessboard, pattern_grid_or_cell_size=cell_size) if world_to_camera_transf is not None: pattern_grid, _ = auxV.get_calib_pattern_info( is_chessboard, cell_size) found, corners = auxV.find_calib_pattern(img, is_chessboard, pattern_grid) cv2.drawChessboardCorners(img, pattern_grid, corners, found) cam_pos = auxV.cam_to_world_coords([0, 0, 0], world_to_camera_transf) d = auxV.world_to_cam_coords([0, 0, 0], world_to_camera_transf)[2] cv2.putText( img, "[x={p[0]:.2f}, y={p[1]:.2f}, z={p[2]:.2f}] cm; dist={d:.2f}cm; F={F:.2f}px" .format(p=100 * cam_pos, d=100 * d, F=F), (50, img.shape[0] - 50), cv2.FONT_HERSHEY_DUPLEX, 1, (0, 0, 200), 1, cv2.LINE_AA) # # Test img->world + world->img transforms to make sure everything works correctly # dist_img_coords = [img.shape[1], img.shape[0], 20] # RADIUS_IN_M = 0.02 # depth = F * RADIUS_IN_M/dist_img_coords[2] # If an object that's dist_img_coords[2] pixels wide measures RADIUS_IN_M meters, how far away from the camera is it? # cam_coords = auxV.img_to_cam_coords(np.hstack((dist_img_coords[0:2], depth)), calibration_file, is_dist_img_coords=True) # world_coords = auxV.cam_to_world_coords(cam_coords, world_to_camera_transf) # new_cam_coords = auxV.world_to_cam_coords(world_coords, world_to_camera_transf) # new_dist_img_coords = auxV.cam_to_img_coords(new_cam_coords, calibration_file, want_dist_img_coords=True) # cv2.putText(img, "[u={p[0]:.2f}, v={p[1]:.2f}]".format(p=new_dist_img_coords), (50, img.shape[0]-100), cv2.FONT_HERSHEY_DUPLEX, 1, (0, 0, 200), 1, cv2.LINE_AA) # cv2.circle(img, tuple([int(x) for x in new_dist_img_coords]), int(dist_img_coords[2]), (0, 200, 0), -1) else: cv2.putText(img, "CAN'T FIND PATTERN!!", (50, img.shape[0] - 50), cv2.FONT_HERSHEY_DUPLEX, 1, (0, 0, 255), 1, cv2.LINE_AA) cv2.imshow(win_title, img) t_end = datetime.now() print("Frame took {:.2f}ms to process!".format( (t_end - t_start).total_seconds() * 1000)) cv2.destroyAllWindows()
class CamFrameGrabberThread(QThread): """ This class implements a background task that constantly grabs new frames from the selected camera """ DEFAULT_CAM = "USB 2.0 Camera" DEFAULT_FPS = 60 FPS_ESTIMATION_ALPHA = 0.98 FRAME_SIZE_STR_SPLITTER = " x " sig_update_cam_ui = pyqtSignal() # Signal that's emitted whenever the UI associated to the camera controls needs to be refreshed (eg: new frame size selected -> need to resize window, error grabbing camera frame -> disconnect camera...) sig_new_dev_selected = pyqtSignal(str) # Signal that's emitted when a device is selected (could be the device that was already selected). It's emitted either from the UI when the user changes the selected item in drpInput, or from the code when sig_new_dev is emitted (as a new device is connected or an old one disconnected) sig_new_img = pyqtSignal() # Signal that's emitted every time a new camera frame becomes available sig_error = pyqtSignal(str) # Signal that's emitted every time an error occurred def __init__(self): QThread.__init__(self) self.done = False self.dev_selected_name = self.DEFAULT_CAM self.cap = None self.qPix = QPixmap(1, 1) self.cv = DroneController() self.cv.load_color_thresh_settings() self.cv.load_blob_detector_settings() self.cv_do_color = True self.cv_do_blob = True self.cv_evaluate_px_at = np.array([0, 0]) self.cv_evaluate_px_value = np.array([0, 0, 0], dtype=np.uint8) self.index = 0 self.actual_fps = 0 def __del__(self): self.wait() @pyqtSlot() def stop_thread(self): # This method is used to externally request this thread to terminate self.done = True # Flag to indicate the outer while loop in run_experiment() to finish thread execution self.change_selected_device("EXIT! :P") # Choose a void device, to force the inner while loop in run_experiment() to exit after grabbing at most one frame if self.wait(5000) is not True or self.cap is not None: # Worst-case scenario: force quit if self.cap is still not None after 5sec logging.warning("Uh-ooohh, CamFrameGrabberThread didn't terminate even after waiting for 5sec. Force quitting :S") self.cap = None self.terminate() else: logging.info("CamFrameGrabberThread exited cleanly :)") @pyqtSlot(str) def change_selected_device(self, dev_selected): if self.dev_selected_name != dev_selected: logging.info("New device selected: {}".format(dev_selected)) self.dev_selected_name = str(dev_selected) @pyqtSlot(str, bool) def change_cam_frame_size(self, new_frame_size, emit_signal): if self.cap is not None and window.drpSize.count() > 0: # Sanity check (in case capture device was just closed or we're clearing drpSize to re-add available sizes) logging.info("Changing frame size to {}".format(new_frame_size)) self.cap.frame_size = tuple([int(x) for x in new_frame_size.split(self.FRAME_SIZE_STR_SPLITTER)]) if emit_signal: # If the user was the one who clicked on this combobox item (as opposed to me, manually from the code), update cam UI self.sig_update_cam_ui.emit() # drpFPS might need to be updated (different frame sizes might have different fps available) @pyqtSlot(str) def change_cam_frame_rate(self, new_frame_rate): if self.cap is not None and window.drpFPS.count() > 0: # Sanity check (in case capture device was just closed or we're clearing drpSize to re-add available sizes) logging.info("Changing frame rate to {}".format(new_frame_rate)) self.cap.frame_rate = int("".join(x for x in new_frame_rate if x.isdigit())) # Keep only the numbers (= remove the " fps" part, if exists) @pyqtSlot(str, int, bool) def change_cam_control_setting(self, ctrl_name, new_value, is_bool): """ Process a request (from the UI) to change a particular camera setting. :param ctrl_name: Particular camera setting that wants to be modified (eg: exposure time, white balance...) :param new_value: New value desired for the camera setting :param is_bool: True if the UI control is a CheckBox, False if it's a SpinBox (ie: TextBox with up&down arrows) """ if self.cap is not None: # Sanity check (in case capture device was just closed) for c in self.cap.controls: if c.display_name == ctrl_name: try: if is_bool: c.set_value(c.min_val if (new_value == Qt.Unchecked) else c.max_val) else: c.set_value(new_value) logging.info("'{}' changed to {}".format(c.display_name, c.value)) except: self.sig_error.emit("Unable to change '{}' property to '{}'! Make sure the value is valid (within bounds, not disabled by other setting like 'Auto' modes, etc).".format(ctrl_name, new_value)) return @pyqtSlot(bool) def set_cv_do_color(self, checked): self.cv_do_color = checked window.grpBlob.setEnabled(checked) self.set_cv_do_blob(False if not checked else window.grpBlob.isChecked()) @pyqtSlot(bool) def set_cv_do_blob(self, checked): self.cv_do_blob = checked @pyqtSlot(int, str, bool) def change_HSV_setting(self, new_value, letter, is_max): """ Updates the corresponding field (H,S or V variable, min or max) of the HSV color thresholding filter. :param new_value: New value to set the corresponding threshold to :param letter: either 'H', 'S', or 'V' (case insensitive), depending on which threshold should be modified :param is_max: True to change the upper threshold; False to change the lower one """ letter = letter.lower() # To avoid problems, convert the letter to lowercase and only compare to lowercase index = 0 if letter == 'h' else 1 if letter == 's' else 2 # Convert H, S, V to 0, 1, 2 # Now update the right variable based on is_max if is_max: self.cv.cv_HSV_thresh_max[index] = new_value else: self.cv.cv_HSV_thresh_min[index] = new_value @pyqtSlot(int, str, bool) def change_blob_setting(self, new_value, letter, is_max): """ Updates the corresponding setting (area, circularity, convexity or inertia, min or max) of the blob detector. :param new_value: If int: New value to set the corresponding threshold to; if bool: activate/deactivate filter :param letter: either 'A', 'C', 'V', or 'I' (case insensitive), depending on which setting should be modified :param is_max: True to change the upper threshold; False to change the lower one [ignored if new_value is bool] """ letter = letter.lower() # To avoid problems, convert the letter to lowercase and only compare to lowercase param = "Area" if letter == 'a' else "Circularity" if letter == 'c' else "Convexity" if letter == 'v' else "Inertia" if type(new_value) is bool: setattr(self.cv.cv_blob_detect_params, "filterBy{}".format(param), new_value) else: setattr(self.cv.cv_blob_detect_params, "{}{}{}".format("max" if is_max else "min", param, "Ratio" if letter == 'i' else ""), new_value) def run(self): while not self.done: self.sleep(1) # Outer while loop just waits until a device is selected self.cap = UvcCapture(self.dev_selected_name) if self.cap is None: # If we didn't find the desired cam, don't continue self.qPix = QPixmap(1, 1) # Default to a black pixel self.cv_evaluate_px_value = np.array([0, 0, 0], dtype=np.uint8) logging.info("No compatible cameras found or chosen camera name not available :(") self.sig_update_cam_ui.emit() # Make sure camera-related UI is updated (eg: don't show exposure time control if no camera is selected...) continue # If we made it here, we successfully connected to the selected camera. Print camera info and select default values for c in self.cap.controls: logging.info("\t{} = {} {} (def:{}, min:{}, max:{})".format(c.display_name, str(c.value), str(c.unit), str(c.def_val), str(c.min_val), str(c.max_val))) self.cap.select_best_frame_mode(self.DEFAULT_FPS) logging.info(self.cap.name + " has the following available modes:\n\t" + str([tuple(x) for x in self.cap.sorted_available_modes()]) + "\nSelected mode: " + str(self.cap.frame_mode)) self.cap.print_info() logging.debug("LOADING SETTINGS returned {}".format(self.cap.load_settings(self.cv.CAMERA_SETTINGS_FILE))) self.sig_update_cam_ui.emit() # Update camera-related UI (eg: add controls for exposure time, white balance, etc.) img = np.zeros(self.cap.frame_size) # Resize qPix to fit new image size self.qPix = QPixmap.fromImage(QImage(img.data, img.shape[1], img.shape[0], QImage.Format_RGB888)) self.actual_fps = self.cap.frame_rate tt = datetime.now() # Initialize tt (used to estimate actual frame rate) to prevent an error on the first loop iteration while self.cap.name == self.dev_selected_name: # Run an inner while loop to capture frames as long as the selected device is kept constant ok = False t = datetime.now() for a in range(5): try: frame = self.cap.get_frame_robust() self.index = frame.index # logging.debug("DEBUG - Successfully got frame {:d} after {:d} tries".format(frame.index, a+1)) ok = True break except uvc.CaptureError as e: logging.error('DEBUG - Could not get Frame. Error: "{}". Tried {} time{}.'.format(e.message, a+1, "s" if a > 0 else "")) if not ok: self.sig_error.emit("Couldn't get camera frame after 5 tries, reconnecting...") break # Exit inner loop (force an iteration on the outer loop) after 5 consecutive failed attempts to grab a frame # We successfully grabbed a frame, let's store the value of the pixel cv_evaluate_px_at for debugging purposes try: self.cv_evaluate_px_value = np.array(frame.bgr[self.cv_evaluate_px_at[1], self.cv_evaluate_px_at[0]]) # Create a new np.array so that the value at the pixel is not cached, but copied (ie: if frame.bgr is modified, cv_evaluate_px_value is not). Also note that images should be indexed [y,x]! except: self.cv_evaluate_px_at = np.array([0, 0]) # Now we can run the CF detection algorithm on it (if grpColor is checked) and modify frame.bgr if we want t2 = datetime.now() if self.cv_do_color: self.cv.detect_cf_in_camera(frame.bgr, self.cv_do_blob) # Run CF detection (and blob detection if checked) # Now overlay the mask over the camera frame t3 = datetime.now() if self.cv.cf_pos_tracked and self.cv_do_blob: cv2.circle(self.cv.cv_cam_frame, tuple(self.cv.cf_curr_pos[0:2].astype(int)), int(self.cv.cf_curr_pos[2]+5), [0, 255, 0], -1) # np.putmask(self.cv.cv_cam_frame[:,:,0], self.cv.cv_filtered_HSV_mask, 255) # np.putmask(self.cv.cv_cam_frame[:,:,1], self.cv.cv_filtered_HSV_mask, 0) # np.putmask(self.cv.cv_cam_frame[:,:,2], self.cv.cv_filtered_HSV_mask, 255) mask = cv2.cvtColor(self.cv.cv_filtered_HSV_mask, cv2.COLOR_GRAY2BGR).astype(bool) np.putmask(self.cv.cv_cam_frame, mask, np.array([255, 0, 255], dtype=np.uint8)) else: t3 = datetime.now() # For consistency, save current time as t3 (even if we didn't do anything # And finally convert the resulting image to qPix and emit the appropriate signal so the UI can refresh it t4 = datetime.now() img = cv2.cvtColor(frame.bgr, cv2.COLOR_BGR2RGB) self.qPix.convertFromImage(QImage(img.data, img.shape[1], img.shape[0], QImage.Format_RGB888)) self.sig_new_img.emit() t5 = datetime.now() self.actual_fps = self.FPS_ESTIMATION_ALPHA*self.actual_fps + (1-self.FPS_ESTIMATION_ALPHA)/(t5-tt).total_seconds() logging.debug("At t={} picture #{:03d} was taken; deltaTtotal={:6.2f}ms [{:6.2f}ms] (capture_frame->{:6.2f}ms + detect->{:6.2f}ms + blob->{:6.2f}ms + putmask->{:6.2f}ms + toQPixmap->{:6.2f}ms) -> Estimated FPS: {:.2f}".format(t2, frame.index, (t5-tt).total_seconds()*1000, (t5-t).total_seconds()*1000, (t2-t).total_seconds()*1000, (self.cv.t_events[-2]-t2).total_seconds()*1000 if self.cv_do_color else 0, (self.cv.t_events[-1]-self.cv.t_events[-2]).total_seconds()*1000 if self.cv_do_blob else 0, (t4-t3).total_seconds()*1000, (t5-t4).total_seconds()*1000, self.actual_fps)) tt = datetime.now() self.cap = None # Set cap to None to automatically release & disconnect from the camera
def run(self): while not self.done: self.sleep(1) # Outer while loop just waits until a device is selected self.cap = UvcCapture(self.dev_selected_name) if self.cap is None: # If we didn't find the desired cam, don't continue self.qPix = QPixmap(1, 1) # Default to a black pixel self.cv_evaluate_px_value = np.array([0, 0, 0], dtype=np.uint8) logging.info("No compatible cameras found or chosen camera name not available :(") self.sig_update_cam_ui.emit() # Make sure camera-related UI is updated (eg: don't show exposure time control if no camera is selected...) continue # If we made it here, we successfully connected to the selected camera. Print camera info and select default values for c in self.cap.controls: logging.info("\t{} = {} {} (def:{}, min:{}, max:{})".format(c.display_name, str(c.value), str(c.unit), str(c.def_val), str(c.min_val), str(c.max_val))) self.cap.select_best_frame_mode(self.DEFAULT_FPS) logging.info(self.cap.name + " has the following available modes:\n\t" + str([tuple(x) for x in self.cap.sorted_available_modes()]) + "\nSelected mode: " + str(self.cap.frame_mode)) self.cap.print_info() logging.debug("LOADING SETTINGS returned {}".format(self.cap.load_settings(self.cv.CAMERA_SETTINGS_FILE))) self.sig_update_cam_ui.emit() # Update camera-related UI (eg: add controls for exposure time, white balance, etc.) img = np.zeros(self.cap.frame_size) # Resize qPix to fit new image size self.qPix = QPixmap.fromImage(QImage(img.data, img.shape[1], img.shape[0], QImage.Format_RGB888)) self.actual_fps = self.cap.frame_rate tt = datetime.now() # Initialize tt (used to estimate actual frame rate) to prevent an error on the first loop iteration while self.cap.name == self.dev_selected_name: # Run an inner while loop to capture frames as long as the selected device is kept constant ok = False t = datetime.now() for a in range(5): try: frame = self.cap.get_frame_robust() self.index = frame.index # logging.debug("DEBUG - Successfully got frame {:d} after {:d} tries".format(frame.index, a+1)) ok = True break except uvc.CaptureError as e: logging.error('DEBUG - Could not get Frame. Error: "{}". Tried {} time{}.'.format(e.message, a+1, "s" if a > 0 else "")) if not ok: self.sig_error.emit("Couldn't get camera frame after 5 tries, reconnecting...") break # Exit inner loop (force an iteration on the outer loop) after 5 consecutive failed attempts to grab a frame # We successfully grabbed a frame, let's store the value of the pixel cv_evaluate_px_at for debugging purposes try: self.cv_evaluate_px_value = np.array(frame.bgr[self.cv_evaluate_px_at[1], self.cv_evaluate_px_at[0]]) # Create a new np.array so that the value at the pixel is not cached, but copied (ie: if frame.bgr is modified, cv_evaluate_px_value is not). Also note that images should be indexed [y,x]! except: self.cv_evaluate_px_at = np.array([0, 0]) # Now we can run the CF detection algorithm on it (if grpColor is checked) and modify frame.bgr if we want t2 = datetime.now() if self.cv_do_color: self.cv.detect_cf_in_camera(frame.bgr, self.cv_do_blob) # Run CF detection (and blob detection if checked) # Now overlay the mask over the camera frame t3 = datetime.now() if self.cv.cf_pos_tracked and self.cv_do_blob: cv2.circle(self.cv.cv_cam_frame, tuple(self.cv.cf_curr_pos[0:2].astype(int)), int(self.cv.cf_curr_pos[2]+5), [0, 255, 0], -1) # np.putmask(self.cv.cv_cam_frame[:,:,0], self.cv.cv_filtered_HSV_mask, 255) # np.putmask(self.cv.cv_cam_frame[:,:,1], self.cv.cv_filtered_HSV_mask, 0) # np.putmask(self.cv.cv_cam_frame[:,:,2], self.cv.cv_filtered_HSV_mask, 255) mask = cv2.cvtColor(self.cv.cv_filtered_HSV_mask, cv2.COLOR_GRAY2BGR).astype(bool) np.putmask(self.cv.cv_cam_frame, mask, np.array([255, 0, 255], dtype=np.uint8)) else: t3 = datetime.now() # For consistency, save current time as t3 (even if we didn't do anything # And finally convert the resulting image to qPix and emit the appropriate signal so the UI can refresh it t4 = datetime.now() img = cv2.cvtColor(frame.bgr, cv2.COLOR_BGR2RGB) self.qPix.convertFromImage(QImage(img.data, img.shape[1], img.shape[0], QImage.Format_RGB888)) self.sig_new_img.emit() t5 = datetime.now() self.actual_fps = self.FPS_ESTIMATION_ALPHA*self.actual_fps + (1-self.FPS_ESTIMATION_ALPHA)/(t5-tt).total_seconds() logging.debug("At t={} picture #{:03d} was taken; deltaTtotal={:6.2f}ms [{:6.2f}ms] (capture_frame->{:6.2f}ms + detect->{:6.2f}ms + blob->{:6.2f}ms + putmask->{:6.2f}ms + toQPixmap->{:6.2f}ms) -> Estimated FPS: {:.2f}".format(t2, frame.index, (t5-tt).total_seconds()*1000, (t5-t).total_seconds()*1000, (t2-t).total_seconds()*1000, (self.cv.t_events[-2]-t2).total_seconds()*1000 if self.cv_do_color else 0, (self.cv.t_events[-1]-self.cv.t_events[-2]).total_seconds()*1000 if self.cv_do_blob else 0, (t4-t3).total_seconds()*1000, (t5-t4).total_seconds()*1000, self.actual_fps)) tt = datetime.now() self.cap = None # Set cap to None to automatically release & disconnect from the camera
class DroneController: COLOR_BALL_TRACKED = (255, 0, 0) COLOR_BALL_UNTRACKED = (0, 0, 255) COLOR_LINE_TRACKED = (255, 0, 0) COLOR_LINE_UNTRACKED = (0, 0, 255) COLOR_TARGET_TRACKED = (0, 255, 0) COLOR_TARGET_UNTRACKED = (0, 0, 255) FIGURE_NAME = "Output" SETTINGS_ENVIRONMENT = "Very bright bedroom" CAMERA_SETTINGS_FILE = "config/cam_settings/Camera settings - USB 2.0 Camera - {}.txt".format(SETTINGS_ENVIRONMENT) COLOR_THRESH_SETTINGS_FILE = "config/color_thresh/Color threshold settings - {}.txt".format(SETTINGS_ENVIRONMENT) BLOB_DETECTOR_SETTINGS_FILE = "config/blob_detector/Blob detector settings.txt" SETTINGS_SEPARATOR = UvcCapture.SETTINGS_SEPARATOR # We save files in a csv type of way ASK_FOR_TARGET_YAW = False TAKEOFF_THRUST = 44000 def __init__(self): self.t_start = self.t_frame = self.t_last_frame = datetime.now() self.t_events = [] self.EXPERIMENT_START_DATETIME = str(self.t_start)[:-7].replace(':', '-') self.VIDEO_FOLDER = "img-ns/{}".format(self.EXPERIMENT_START_DATETIME) self.experiment_log = plot_tools.ExperimentLog(self.EXPERIMENT_START_DATETIME, {"Roll": "piv", "Pitch": "piv", "Yaw": "pid", "Thrust": "piv", "Estimated_Z": "log", "Velocity_Z": "log"}) self.window_for_kb_input = None self.video_capture = None self.cv_HSV_thresh_min = np.array([ 0, 0, 0], dtype=np.uint8) self.cv_HSV_thresh_max = np.array([255, 255, 255], dtype=np.uint8) self.cv_blob_detect_params = None self.cv_cam_frame = None self.cv_filtered_HSV_mask = None self.cv_frame_out = None self.crazyflie = None self.cf_log = None self.cf_radio_connecting = True self.cf_radio_connected = False self.cf_ignore_camera = False self.cf_pos_tracked = False self.cf_taking_off = True self.cf_str_status = "TAKING OFF" self.cf_roll = self.cf_pitch = self.cf_yaw = self.cf_estimated_z = self.cf_vel_z = 0 self.cf_curr_pos = np.array([0, 0, 0]) self.cf_PID_roll = PID.PIDposAndVel(posP=0.5, velP=0.05, velI=0.01, vel_offs=0, pos_out_max=300, vel_out_max=30, vel_invert_error=True) self.cf_PID_pitch = PID.PIDposAndVel(posP=0.7, velP=0.3, velI=0.002, vel_offs=0, pos_out_max=30, vel_out_max=30) self.cf_PID_yaw = PID.PID(P=0.5, I=0.3, D=0, offs=0, out_max=20, invert_error=True, error_in_degrees=True) self.cf_PID_thrust = PID.PIDposAndVel(posP=1, velP=35, velI=25, vel_offs=43000, pos_out_max=300, vel_out_max=7000, pos_invert_error=True, vel_invert_input=True) def init_video_cam_and_cv_algorithm(self, create_video_folder=True): """ Initializes camera: connects to it, loads settings from config file (if available), loads color threshold and blob detector settings (if available), creates a folder to save cv output images... :param create_video_folder: True to create folder specified by VIDEO_FOLDER (where output frames will be saved) """ self.video_capture = UvcCapture.new_from_settings(self.CAMERA_SETTINGS_FILE) # Connect to device specified by settings, and load its desired param values if self.video_capture is None: # If unable to connect to device specified by settings, open first available camera self.video_capture = UvcCapture(0) if self.video_capture is None: # If still unable to connect, raise an exception raise Exception("Couldn't open camera! :(") # If we're here, we couldn't connect to device specified by settings but were able to open 1st available cam if not self.video_capture.load_settings(self.CAMERA_SETTINGS_FILE): # Try to load frame size & rate from settings self.video_capture.select_best_frame_mode(60) # If loading settings failed, choose best frame size with fps >= 60 logging.info("Camera opened! :)") # Initialize cv algorithm too: load settings for color thresholding and blob detector self.load_color_thresh_settings() self.load_blob_detector_settings() # Sometimes, first couple frames take a long time to be obtained, do it before quad goes in flying mode self.video_capture.get_frame_robust() self.video_capture.get_frame_robust() # Initialize PID setpoints and initial input value to the center of the frame self.cf_PID_roll.setSetPoint(self.video_capture.frame_size[0] / 2) self.cf_PID_thrust.setSetPoint(self.video_capture.frame_size[1] / 2) self.cf_PID_pitch.setSetPoint(40) self.cf_PID_roll.PIDpos.curr_input = self.cf_PID_roll.getSetPoint() self.cf_PID_thrust.PIDpos.curr_input = self.cf_PID_thrust.getSetPoint() self.cf_PID_pitch.PIDpos.curr_input = self.cf_PID_pitch.getSetPoint() # Prepare the folder self.VIDEO_FOLDER so we can store each frame we processed (for debugging) if create_video_folder: shutil.rmtree(self.VIDEO_FOLDER, ignore_errors=True) # Delete the folder and its contents, if it exists (ignore errors if it doesn't) os.makedirs(self.VIDEO_FOLDER) # Now create the folder, which won't throw any exceptions as we made sure it didn't already exist def init_UI_window(self): """ Initializes the appropriate user interface depending on experiment's purpose (see is_input_for_drone_commands) :param is_input_for_drone_commands: True if we have communication with the CF and need input to send it commands False if we're just debugging the computer vision side (camera settings, color threshold, etc.) and need input to debug pixel information """ # Open an SDL2 window to track keyboard keydown events and use it to send commands to the CF sdl2.ext.init() self.window_for_kb_input = sdl2.ext.Window("Window to receive keyboard input", size=(400, 300)) self.window_for_kb_input.show() def save_color_thresh_settings(self, HSV_min, HSV_max, file_name=COLOR_THRESH_SETTINGS_FILE, sep=SETTINGS_SEPARATOR): """ Saves specified color threshold settings to a file. :param HSV_min: np.array (1x3) containing minimum H, S and V values for the color thresholding CF detection :param HSV_max: np.array (1x3) containing maximum H, S and V values for the color thresholding CF detection :param file_name: Name of the file to use when saving the settings :param sep: String that will be used to separate parameters (in a csv fashion). Eg: '\t', ',', etc. :return: True if everything went well; False if settings couldn't be saved """ try: logging.debug("Saving current color threshold settings to file '{}'".format(file_name)) with open(file_name, 'w') as f: # Open file for output f.write("{}\n".format(sep.join(HSV_min.astype(str)))) # Store HSV_min f.write("{}\n".format(sep.join(HSV_max.astype(str)))) # Store HSV_max except: logging.exception("Something went wrong while saving current color threshold settings to '{}'.".format(file_name)) return False return True def load_color_thresh_settings(self, file_name=COLOR_THRESH_SETTINGS_FILE, sep=SETTINGS_SEPARATOR): """ Loads color threshold settings from a file. :param file_name: Path to the file to load the settings from :param sep: String that was used to separate parameters (in a csv fashion). Eg: '\t', ',', etc. :return: True if everything went well; False if settings couldn't be loaded """ try: logging.debug("Loading color threshold settings from file '{}'".format(file_name)) with open(file_name, 'r') as f: # Open file for input self.cv_HSV_thresh_min = np.array(f.readline().rstrip('\r\n').split(sep), dtype=np.uint8) self.cv_HSV_thresh_max = np.array(f.readline().rstrip('\r\n').split(sep), dtype=np.uint8) logging.debug("\tLoaded color threshold settings: HSV_min={}; HSV_max={}".format(self.cv_HSV_thresh_min, self.cv_HSV_thresh_max)) except: logging.exception("Something went wrong while loading color threshold settings from '{}'.".format(file_name)) return False return True def save_blob_detector_settings(self, detector_params, file_name=BLOB_DETECTOR_SETTINGS_FILE, sep=SETTINGS_SEPARATOR): """ Saves specified blob detector settings to a file. :param detector_params: cv2.SimpleBlobDetector_Params object containing the params which want to be saved :param file_name: Name of the file to use when saving the settings :param sep: String that will be used to separate parameters (in a csv fashion). Eg: '\t', ',', etc. :return: True if everything went well; False if settings couldn't be saved """ try: logging.debug("Saving current blob detector settings to file '{}'".format(file_name)) with open(file_name, 'w') as f: # Open file for output for m in dir(detector_params): # Traverse all methods and properties of detector_params if m[0] != '_': # For every property that's not "built-in" (ie: doesn't start by '_') f.write("{}{}{}\n".format(m, sep, getattr(detector_params, m))) # Store the property name and value except: logging.exception("Something went wrong while saving current blob detector settings to '{}'.".format(file_name)) return False return True def load_blob_detector_settings(self, file_name=BLOB_DETECTOR_SETTINGS_FILE, sep=SETTINGS_SEPARATOR): """ Loads blob detector settings from a file. Leave file_name empty to only load default params. cv2 will use these params to detect the drone from a binary image mask (the output of the color thresholding step). :param file_name: Path to the file to load the settings from :param sep: String that was used to separate parameters (in a csv fashion). Eg: '\t', ',', etc. :return: True if everything went well; False if settings couldn't be loaded """ detector_params = cv2.SimpleBlobDetector_Params() self.cv_blob_detect_params = detector_params # Filter by color detector_params.filterByColor = False detector_params.blobColor = 255 # Change thresholds detector_params.minThreshold = 254 detector_params.maxThreshold = 255 # Filter by Area. detector_params.filterByArea = True detector_params.minArea = 30 detector_params.maxArea = 40000 # Filter by Circularity detector_params.filterByCircularity = False detector_params.minCircularity = 0.7 detector_params.maxCircularity = 1.0 # Filter by Convexity detector_params.filterByConvexity = False detector_params.minConvexity = 0.7 detector_params.maxConvexity = 1.0 # Filter by Inertia detector_params.filterByInertia = False detector_params.minInertiaRatio = 0.5 detector_params.maxInertiaRatio = 1.0 detector_params.minRepeatability = 1 detector_params.minDistBetweenBlobs = 3000 try: logging.debug("Loading blob detector settings from file '{}'".format(file_name)) with open(file_name, 'r') as f: # Open file for input for line in f: # Every line contains one property: name + sep + value name, value = line.split(sep) setattr(detector_params, name, eval(value)) # Use eval to cast to right type (eg: False instead of "False") logging.debug("\tLoaded blob detector setting: '{}' = {}".format(name, value)) except: logging.exception("Something went wrong while loading blob detector settings from '{}'.".format(file_name)) return False return True def connect_to_cf(self, retry_after=10, max_timeout=20): """ Initializes radio drivers, looks for available CrazyRadios, and attempts to connect to the first available one. Doesn't return anything, but raises exceptions if anything goes wrong. :param retry_after: Time in seconds after which we should abort current connection and restart the attempt. :param max_timeout: Time in seconds after which we should give up if we haven't been able to establish comm. yet """ logging.info("Initializing drivers.") crtp.init_drivers(enable_debug_driver=False) logging.info("Setting up the communication link. Looking for available CrazyRadios in range.") available_links = crtp.scan_interfaces() if len(available_links) == 0: raise Exception("Error, no Crazyflies found. Exiting.") else: logging.info("CrazyFlies found:") # For debugging purposes, show info about available links for i in available_links: logging.info("\t" + i[0]) link_uri = available_links[0][0] # Choose first available link logging.info("Initializing CrazyFlie (connecting to first available interface: '{}').".format(link_uri)) self.crazyflie = Crazyflie(ro_cache="cachero", rw_cache="cacherw") # Create an instance of Crazyflie self.crazyflie.connected.add_callback(self.on_cf_radio_connected) # Set up callback functions for communication feedback self.crazyflie.disconnected.add_callback(self.on_cf_radio_disconnected) self.crazyflie.connection_failed.add_callback(self.on_cf_radio_conn_failed) self.crazyflie.connection_lost.add_callback(self.on_cf_radio_conn_lost) cnt = 0 # Initialize a time counter while self.cf_radio_connecting and cnt < max_timeout: if cnt % retry_after == 0: if cnt > 0: # Only show warning after first failed attempt logging.warning("Unable to establish communication with Crazyflie ({}) after {}s. Retrying...".format(link_uri, retry_after)) self.crazyflie.close_link() # Closing the link will call on_disconnect, which will set cf_radio_connecting to False self.cf_radio_connecting = True # Reset cf_radio_connecting back to True self.crazyflie.open_link(link_uri) # Connect/Reconnect to the CrazyRadio through the selected interface time.sleep(1) # Sleep for a second (give time for callback functions to detect whether we are connected) cnt += 1 # Increase the "waiting seconds" counter if cnt >= max_timeout: self.crazyflie.close_link() raise Exception("Unable to establish communication with CrazyFlie after {}s. Given up :(".format(max_timeout)) elif not self.cf_radio_connected: raise Exception("Something failed while attempting to connect to the CrazyFlie, exiting.") # self.crazyflie.commander.send_setpoint(0, 0, 0, 0) # If we successfully connected to the CF, send thrust=0 (new firmware initializes thrustLock=True, only way to unlock it so it executes commands is by setting thrust=0) def setup_cf(self): """ Sets up the CrazyFlie before running the experiment. This includes configuring some params to default values and requesting the CrazyFlie to log certain variables back at constant intervals. Doesn't return anything, but raises exceptions if anything goes wrong. """ try: # Send some default values for CF params self.crazyflie.param.set_value('controller.tiltComp', '{:d}'.format(True)) self.crazyflie.param.set_value('flightmode.poshold', '{:d}'.format(False)) # Disable poshold and althold by default self.crazyflie.param.set_value('flightmode.althold', '{:d}'.format(False)) self.crazyflie.param.set_value('flightmode.posSet', '{:d}'.format(False)) self.crazyflie.param.set_value('flightmode.yawMode', '0') self.crazyflie.param.set_value('flightmode.timeoutStab', '{:d}'.format(1000*60*10)) # Stabilize (rpy=0) CF if doesn't receive a radio command in 10min self.crazyflie.param.set_value('flightmode.timeoutShut', '{:d}'.format(1000*60*20)) # Shutdown CF if doesn't receive a radio command in 20min self.crazyflie.param.set_value('posCtlPid.thrustBase', '{}'.format(self.TAKEOFF_THRUST)) self.crazyflie.param.set_value("ring.effect", "1") # Turn off LED ring self.crazyflie.param.set_value("ring.headlightEnable", "0") # Turn off LED headlight except Exception as e: raise Exception("Couldn't initialize CrazyFlie params to their desired values. Details: {}".format(e.message)) # Create a log configuration and include all variables that want to be logged self.cf_log = LogConfig(name="cf_log", period_in_ms=10) self.cf_log.add_variable("stabilizer.roll", "float") self.cf_log.add_variable("stabilizer.pitch", "float") self.cf_log.add_variable("stabilizer.yaw", "float") self.cf_log.add_variable("posEstimatorAlt.estimatedZ", "float") self.cf_log.add_variable("posEstimatorAlt.velocityZ", "float") try: self.crazyflie.log.add_config(self.cf_log) # Validate the log configuration and attach it to our CF except Exception as e: raise Exception("Couldn't attach the log config to the CrazyFlie, bad configuration. Details: {}".format(e.message)) self.cf_log.data_received_cb.add_callback(self.on_cf_log_new_data) # Register appropriate callbacks self.cf_log.error_cb.add_callback(self.on_cf_log_error) self.cf_log.start() # Start logging! # Get the CF's initial yaw (should be facing the camera) so we can have PID_yaw maintain that orientation if self.ASK_FOR_TARGET_YAW: # Either ask the user to press Enter to indicate the CF's orientation is ready raw_input("\nRotate the drone so it faces the camera, press Enter when you're ready...\n") else: # Or automatically detect the first yaw log packet and set the current orientation as the desired yaw while abs(self.cf_yaw) < 0.01: # Wait until first cf_yaw value is received (cf_yaw=0 by default) time.sleep(0.1) self.cf_PID_yaw.SetPoint = self.cf_yaw print "Target yaw set at {:.2f}.".format(self.cf_yaw) self.crazyflie.add_port_callback(CRTPPort.CONSOLE, self.print_cf_console) self.crazyflie.commander.send_setpoint(0, 0, 0, 0) # New firmware version requires to send thrust=0 at least once to "unlock thrust" def fly_cf(self): """ Provides the structure to make the drone fly (actual flight control is done in hover() though). hover() is called until user stops the experiment, then hover is called for a few more secs to record more data (usually includes information about the CF crashing...) :return: Whether or not to keep (store) the logs, based on whether the CF ever actually took off and flew """ print "Giving you 5s to prepare for take-off..." time.sleep(5) # t = Timer(20, self.m_CrazyFlie.close_link) # Start a timer to automatically disconnect in 20s # t.start() # Prepare for take off: clear PIDs, log start time... self.cf_str_status = "TAKING OFF" self.t_start = datetime.now() self.cf_PID_roll.clear() self.cf_PID_pitch.clear() self.cf_PID_yaw.clear() self.cf_PID_thrust.clear() # Alright, let's fly! tStop = None while tStop is None: # tStop will remain None while everything's fine tStop = self.hover() # hover returns None while everything's fine; the time to end the experiment otherwise # If we get here, either the user stopped the experiment or the code detected something went wrong print "AT t={}, A KEY WAS PRESSED -> STOPPING!".format(datetime.now().strftime("%H:%M:%S.%f")[:-3]) save_logs = (self.cf_str_status != "TAKING OFF") # Only store the logs if the drone ever started flying (not just took off) self.cf_str_status = "STOPPED" # Updated new drone status while datetime.now() < tStop: # Keep calling hover until tStop, so data is still logged self.hover() return save_logs # Return whether or not to keep the logs def cleanup_experiment(self, save_logs=True): """ "Cleans up" the experiment: closes any open windows, saves logs, disconnects camera and drone, etc. """ # If we get here, either the user ended the experiment, or an exception occurred. Same action regardless: if self.cf_log is not None: # If we were logging data from the CF, stop it (will reconnect faster next time) self.cf_log.stop() self.cf_log.delete() if self.crazyflie is not None: # If the CF was ever initialized, make sure the communication link is closed self.crazyflie.close_link() self.video_capture = None # Destroy the video capture object (this takes care of closing the camera etc.) cv2.destroyAllWindows() # Close all UI windows that could be open if self.window_for_kb_input is not None: self.window_for_kb_input.hide() sdl2.ext.quit() if save_logs: # If the experiment didn't crash before starting (the CF ever took off), plot and save the logs self.experiment_log.plot(False) self.experiment_log.save() else: # Otherwise just delete the folder (and its contents) where cam frames would have been/were saved shutil.rmtree(self.VIDEO_FOLDER, ignore_errors=False) def run_experiment(self): """ DroneController's main method: initializes all components (vision, communication, drone params, etc.) then runs the experiment. """ logging.disable(logging.DEBUG) # Seems to work better than .basicConfig(INFO), especially if logging has already been initialized -> Only show messages of level INFO or higher save_logs = True # Use this auxiliary variable to prevent saving logs if the drone never took off try: self.connect_to_cf() # Connect to the CrazyFlie self.setup_cf() # Can't initialize LogConfig until we're connected, because it needs to check that the variables we'd like to add are in the TOC. So this function must be called after connect_to_cf() self.init_video_cam_and_cv_algorithm() # Connect to the first available camera, load default settings, etc. self.init_UI_window() # Open a window to receive user input to control the CF save_logs = self.fly_cf() # And finally fly the CF except: logging.exception("Shutting down due to an exception =( See details below:") # If we got here, either the user ended the experiment, or an exception occurred. Same action regardless: self.cleanup_experiment(save_logs) # "Cleanup" the experiment: close windows, save logs, etc. def get_cf_target_pos(self): """ :return: np.array containing the current (estimated) 3D position of the CrazyFlie """ return np.array([int(round(x)) for x in [self.cf_PID_roll.getSetPoint(), self.cf_PID_thrust.getSetPoint(), self.cf_PID_pitch.getSetPoint()]]) def get_cf_curr_pos(self): """ :return: np.array containing the current (estimated) 3D position of the CrazyFlie """ return np.array([self.cf_PID_roll.getCurrPos(), self.cf_PID_thrust.getCurrPos(), self.cf_PID_pitch.getCurrPos()]) def get_cf_curr_vel(self): """ :return: np.array containing the current (estimated) 3D velocity of the CrazyFlie """ return np.array([self.cf_PID_roll.getCurrVel(), self.cf_PID_thrust.getCurrVel(), self.cf_PID_pitch.getCurrVel()]) def on_cf_radio_connected(self, link_uri): logging.info("Successfully connected to Crazyflie at '{}'!".format(link_uri)) self.cf_radio_connecting = False self.cf_radio_connected = True def on_cf_radio_conn_failed(self, link_uri, msg): logging.error("Connection to '{}' failed: {}.".format(link_uri, msg)) # Initial connection fails (i.e no Crazyflie at the speficied address) self.cf_radio_connecting = False self.cf_radio_connected = False def on_cf_radio_conn_lost(self, link_uri, msg): logging.warning("Connection to '{}' lost: {}.".format(link_uri, msg)) # Disconnected after a connection has been made (i.e Crazyflie moves out of range) def on_cf_radio_disconnected(self, link_uri): logging.error("Disconnected from '{}'.".format(link_uri)) # Crazyflie is disconnected (called in all cases) self.cf_radio_connecting = False self.cf_radio_connected = False def on_cf_log_error(self, logconf, msg): logging.error("Error when logging %s: %s." % (logconf.name, msg)) def on_cf_log_new_data(self, timestamp, data, logconf): logging.debug("[%d][%s]: %s" % (timestamp, logconf.name, data)) self.cf_roll = data['stabilizer.roll'] self.cf_pitch = data['stabilizer.pitch'] self.cf_yaw = data['stabilizer.yaw'] self.cf_estimated_z = data['posEstimatorAlt.estimatedZ'] self.cf_vel_z = data['posEstimatorAlt.velocityZ'] print "\rCurrent yaw: {:.2f}deg".format(self.cf_yaw), def print_cf_console(self, packet): console_text = packet.data.decode('UTF-8') print("Console: {}".format(console_text)) def send_cf_param(self, complete_name, value): """ Modified version of crazyflie.param.set_value that sends the packet immediately (instead of using a Thread+Queue """ element = self.crazyflie.param.toc.get_element_by_complete_name(complete_name) if not element: raise KeyError("Couldn't set {}={}, param is not in the TOC!".format(complete_name, value)) elif element.access == ParamTocElement.RO_ACCESS: raise AttributeError("Couldn't set {}={}, param is read-only!".format(complete_name, value)) else: varid = element.ident pk = CRTPPacket() pk.set_header(CRTPPort.PARAM, WRITE_CHANNEL) pk.data = struct.pack('<B', varid) pk.data += struct.pack(element.pytype, eval(value)) self.crazyflie.send_packet(pk, expected_reply=(tuple(pk.data[0:2]))) def process_kb_input(self): """ Processes all keydown events, and takes the corresponding action depending on which key was pressed. :return: Boolean indicating whether the experiment shall go on: True while everything's fine, False to stop it """ events = sdl2.ext.get_events() # Fetch any new input event for event in events: # Iterate through all of them if event.type == sdl2.SDL_KEYDOWN: # And only process keydown events key_orig = event.key.keysym.sym # Fetch the key that was pressed down try: key = chr(key_orig) # Try to convert the key to char except: # Usually, this exeption occurs when a key can't be converted to char (arrows, Esc, etc.) if key_orig == sdl2.SDLK_UP: key = "Up" elif key_orig == sdl2.SDLK_DOWN: key = "Down" elif key_orig == sdl2.SDLK_ESCAPE: key = "Esc" else: key = hex(key_orig) logging.info("Key: '{}'".format(key)) self.window_for_kb_input.title = "Last key pressed: '{}' at t={}".format(key, str(datetime.now().time())[:-3]) key = key.lower() # Convert to lowercase so we don't have to worry about different cases if key == 'a': # Move left self.cf_PID_roll.setSetPoint(self.cf_PID_roll.getSetPoint() + 20) elif key == 's': # Move back self.cf_PID_pitch.setSetPoint(max(self.cf_PID_pitch.getSetPoint() - 2, 1)) # Size (ball radius) can't be negative! Make sure depth value is at least 1px elif key == 'd': # Move right self.cf_PID_roll.setSetPoint(self.cf_PID_roll.getSetPoint() - 20) elif key == 'w': # Move forward self.cf_PID_pitch.setSetPoint(self.cf_PID_pitch.getSetPoint() + 2) elif key == 'u': # Move up self.cf_PID_thrust.setSetPoint(self.cf_PID_thrust.getSetPoint() - 20) elif key == 'h': # Move down self.cf_PID_thrust.setSetPoint(self.cf_PID_thrust.getSetPoint() + 20) elif key == 'f': # Toggle altitude hold mode if self.cf_ignore_camera: self.cf_PID_thrust.clear() # If we were ignoring the camera for Z, thrust PID will have a wrong I component self.cf_ignore_camera = not self.cf_ignore_camera self.cf_str_status = "NO CAM for Z" if self.cf_ignore_camera else "FULL CAM" # self.crazyflie.param.set_value('flightmode.althold', '{:d}'.format(self.cf_ignore_camera)) # while not self.crazyflie.param.param_updater.request_queue.empty(): # Wait for the packet to be sent # time.sleep(0.01) self.send_cf_param('flightmode.althold', '{:d}'.format(self.cf_ignore_camera)) elif key == 'e': # Stop taking off and start flying (hover at current position) self.cf_taking_off = False self.cf_str_status = "FLYING" self.cf_PID_roll.setSetPoint(self.cf_PID_roll.getCurrPos()) self.cf_PID_pitch.setSetPoint(self.cf_PID_pitch.getCurrPos() + 2) self.cf_PID_thrust.setSetPoint(self.cf_PID_thrust.getCurrPos()) self.cf_PID_roll.clear() self.cf_PID_pitch.clear() self.cf_PID_thrust.clear() else: # Any other key ends the experiment # self.crazyflie.param.set_value('flightmode.althold', '{:d}'.format(False)) # Make sure we're not on althold mode, so sending a thrust 0 will kill the motors and not just descend self.send_cf_param('flightmode.althold', '{:d}'.format(False)) # Make sure we're not on althold mode, so sending a thrust 0 will kill the motors and not just descend return False return True def detect_cf_in_camera(self, frame=None, find_blob=True): """ Runs an iteration of the computer vision algorithm that estimates the CrazyFlie's position in 3D. That is, it captures a frame from the camera, converts it to HSV, filters by color, and detects a blob. Saves the 3D position in cf_curr_pos, camera frame in cv_cam_frame and color filter mask in cv_filtered_HSV_mask. """ ######################################################### # CAPTURE FRAME # ######################################################### self.t_events = [datetime.now()] self.cv_cam_frame = frame # Allow the function to receive a frame if we already got it from the camera if self.cv_cam_frame is None: # Otherwise, if frame is None (default), grab a frame from the camera try: uvc_frame = self.video_capture.get_frame_robust() # read() blocks execution until a new frame arrives! -> Obtain t AFTER grabbing the frame self.t_events.append(datetime.now()) self.t_frame = self.t_events[-1] self.cv_cam_frame = uvc_frame.bgr # .copy() self.t_events.append(datetime.now()) except Exception as e: raise Exception("Unexpected error accessing the camera frame :( Details: {}.".format(e.message)) ######################################################### # COLOR THRESHOLD # ######################################################### # self.cv_cam_frame = cv2.resize(self.cv_cam_frame, None, fx=0.5, fy=0.5) # self.t_events.append(datetime.now()) # self.cv_cam_frame = cv2.GaussianBlur(self.cv_cam_frame, (3, 3), 0) # Not needed, camera is already physically "blurring" (sharpness parameter set to 0) # self.t_events.append(datetime.now()) frame_HSV = cv2.cvtColor(self.cv_cam_frame, cv2.COLOR_BGR2HSV) self.t_events.append(datetime.now()) self.cv_filtered_HSV_mask = cv2.inRange(frame_HSV, self.cv_HSV_thresh_min, self.cv_HSV_thresh_max) self.t_events.append(datetime.now()) self.cv_filtered_HSV_mask = cv2.morphologyEx(self.cv_filtered_HSV_mask, cv2.MORPH_OPEN, cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (5, 5))) self.t_events.append(datetime.now()) ######################################################### # DRONE DETECTION # ######################################################### if find_blob: keypoints = cv2.SimpleBlobDetector_create(self.cv_blob_detect_params).detect(self.cv_filtered_HSV_mask) self.t_events.append(datetime.now()) self.cf_pos_tracked = bool(keypoints) # For now we determine the CF's position is tracked if we find at least 1 blob if keypoints: # If the cv algorithm detected at least one blob keypoint = max(keypoints, key=attrgetter('size')) # Focus on the biggest blob self.cf_curr_pos = np.hstack((keypoint.pt, keypoint.size/2)) # And save the position estimated by the CV algorithm def control_cf(self): """ Sends messages to control the CrazyFlie (roll-pitch-yaw-thrust setpoints) using the current location and PID loops. :param drone_curr_pos: 3D np.array containing the current (estimated) position of the drone """ if not self.cf_pos_tracked: # If cv algorithm wasn't able to detect the drone, linearly estimate its position based on previous position and speed # Since PIDs haven't been updated with current values yet, don't have to multiply velocity by (curr_time-last_time) but rather by (t_frame-curr_time) self.cf_curr_pos = \ np.array([self.cf_PID_roll.getCurrPos(), self.cf_PID_thrust.getCurrPos(), self.cf_PID_pitch.getCurrPos()]) + \ np.array([self.cf_PID_roll.getCurrVel(), -self.cf_PID_thrust.getCurrVel(), self.cf_PID_pitch.getCurrVel()]) * \ (self.t_frame - self.cf_PID_thrust.PIDvel.curr_time).total_seconds() self.cf_curr_pos[2] = max(1, self.cf_curr_pos[2]) # Make sure 3D (size) stays positive! (Prevents errors further down the line) # Update PID loops with new position at t=self.t_frame self.cf_PID_roll.update(self.cf_curr_pos[0], self.t_frame) self.cf_PID_pitch.update(self.cf_curr_pos[2], self.t_frame) self.cf_PID_yaw.update(self.cf_yaw, self.t_frame) self.cf_PID_thrust.update(self.cf_curr_pos[1], self.t_frame) # Log all relevant variables after each iteration self.experiment_log.update(roll=self.cf_PID_roll, pitch=self.cf_PID_pitch, yaw=self.cf_PID_yaw, thrust=self.cf_PID_thrust, estimated_z=self.cf_estimated_z, velocity_z=self.cf_vel_z) # Send the appropriate roll-pitch-yaw-thrust setpoint depending on the scenario (eg: taking off, stopping, etc.) if self.cf_radio_connected: # While the experiment is running if self.cf_taking_off: # If taking off, send a constant RPYT setpoint that will make the CF go up "straight" # self.crazyflie.commander.send_setpoint(self.cf_PID_roll.PIDvel.out_offs, self.cf_PID_pitch.PIDvel.out_offs, self.cf_PID_yaw.output, self.TAKEOFF_THRUST) self.crazyflie.commander.send_setpoint(self.cf_PID_roll.PIDvel.out_offs, self.cf_PID_pitch.PIDvel.out_offs, 0, self.TAKEOFF_THRUST) else: # This condition holds ever after take off (once the user presses the key to start "flying") if self.cf_ignore_camera: # If user selected it, control the drone in althold mode (altitude is done on-board, rest is still controlled with the cam) # self.crazyflie.commander.send_setpoint(self.cf_PID_roll.getOutput(), self.cf_PID_pitch.getOutput(), self.cf_PID_yaw.output, 32767) self.crazyflie.commander.send_setpoint(self.cf_PID_roll.getOutput(), self.cf_PID_pitch.getOutput(), 0, 32767) elif not self.cf_pos_tracked: # If we couldn't find the drone, don't send any commands (not to mislead it) pass else: # Otherwise, just use the camera to control all 4 independent axes # self.crazyflie.commander.send_setpoint(self.cf_PID_roll.getOutput(), self.cf_PID_pitch.getOutput(), self.cf_PID_yaw.output, self.cf_PID_thrust.getOutput()) self.crazyflie.commander.send_setpoint(self.cf_PID_roll.getOutput(), self.cf_PID_pitch.getOutput(), 0, self.cf_PID_thrust.getOutput()) else: # If the user has decided to end the experiment, kill the motors and reset PIDs (this is not really necessary) self.crazyflie.commander.send_setpoint(0, 0, 0, 0) self.cf_PID_roll.clear() self.cf_PID_pitch.clear() self.cf_PID_yaw.clear() self.cf_PID_thrust.clear() self.t_events.append(datetime.now()) def get_OSD_text(self, t): """ Generates written debug information (OSD=on-screen display) to be displayed at the bottom of the current frame :param t: Datetime at which camera frame was obtained (through datetime.now()) :return: String containing relevant debug information (PID values, drone estimated position, etc.) """ formatNum = "{:+6.2f}" strPrint = ("ROLLp. ={:+3.0f};" + formatNum + " [" + formatNum + "," + formatNum + "," + formatNum + "]\t\t" + "ROLLv. ={:+3.0f};" + formatNum + " [" + formatNum + "," + formatNum + "," + formatNum + "]\t\t" + "PITCHp={:+3.0f};" + formatNum + " [" + formatNum + "," + formatNum + "," + formatNum + "]\t\t" + "PITCHv={:+3.0f};" + formatNum + " [" + formatNum + "," + formatNum + "," + formatNum + "]\t\t" + "YAW.. ={:+3.0f};" + formatNum + " [" + formatNum + "," + formatNum + "," + formatNum + "]\t\t" + "THRUSp={:3.0f};{:6.0f} [{:+6.0f}, {:+6.0f}, {:+6.0f}]\t\t" + "THRUSv={:3.0f};{:6.0f} [{:+6.0f}, {:+6.0f}, {:+6.0f}]\t\t" + "[x:{:4.0f}, y:{:4.0f}, z:{:4.0f}], [vx:{:4.0f}, vy:{:4.0f}, vz:{:4.0f}], " + "rpy: " + formatNum + "," + formatNum + "," + formatNum + "]\t\t" + "@{} (FPS: {:5.2f}) - " + self.cf_str_status).format( self.cf_PID_roll.PIDpos.SetPoint, self.cf_PID_roll.PIDpos.output, self.cf_PID_roll.PIDpos.PTerm, self.cf_PID_roll.PIDpos.Ki * self.cf_PID_roll.PIDpos.ITerm, self.cf_PID_roll.PIDpos.Kd * self.cf_PID_roll.PIDpos.DTerm, self.cf_PID_roll.PIDvel.SetPoint, self.cf_PID_roll.PIDvel.output, self.cf_PID_roll.PIDvel.PTerm, self.cf_PID_roll.PIDvel.Ki * self.cf_PID_roll.PIDvel.ITerm, self.cf_PID_roll.PIDvel.Kd * self.cf_PID_roll.PIDvel.DTerm, self.cf_PID_pitch.PIDpos.SetPoint, self.cf_PID_pitch.PIDpos.output, self.cf_PID_pitch.PIDpos.PTerm, self.cf_PID_pitch.PIDpos.Ki * self.cf_PID_pitch.PIDpos.ITerm, self.cf_PID_pitch.PIDpos.Kd * self.cf_PID_pitch.PIDpos.DTerm, self.cf_PID_pitch.PIDvel.SetPoint, self.cf_PID_pitch.PIDvel.output, self.cf_PID_pitch.PIDvel.PTerm, self.cf_PID_pitch.PIDvel.Ki * self.cf_PID_pitch.PIDvel.ITerm, self.cf_PID_pitch.PIDvel.Kd * self.cf_PID_pitch.PIDvel.DTerm, self.cf_PID_yaw.SetPoint, self.cf_PID_yaw.output, self.cf_PID_yaw.PTerm, self.cf_PID_yaw.Ki * self.cf_PID_yaw.ITerm, self.cf_PID_yaw.Kd * self.cf_PID_yaw.DTerm, self.cf_PID_thrust.PIDpos.SetPoint, self.cf_PID_thrust.PIDpos.output, self.cf_PID_thrust.PIDpos.PTerm, self.cf_PID_thrust.PIDpos.Ki * self.cf_PID_thrust.PIDpos.ITerm, self.cf_PID_thrust.PIDpos.Kd * self.cf_PID_thrust.PIDpos.DTerm, self.cf_PID_thrust.PIDvel.SetPoint, self.cf_PID_thrust.PIDvel.output, self.cf_PID_thrust.PIDvel.PTerm, self.cf_PID_thrust.PIDvel.Ki * self.cf_PID_thrust.PIDvel.ITerm, self.cf_PID_thrust.PIDvel.Kd * self.cf_PID_thrust.PIDvel.DTerm, self.cf_PID_roll.getCurrPos(), self.cf_PID_thrust.getCurrPos(), self.cf_PID_pitch.getCurrPos(), self.cf_PID_roll.getCurrVel(), self.cf_PID_thrust.getCurrVel(), self.cf_PID_pitch.getCurrVel(), self.cf_roll, self.cf_pitch, self.cf_yaw, str(t-self.t_start)[3:-3], 1./(t-self.t_last_frame).total_seconds()) # logging.debug(strPrint) return " SP | SENT [ P , I , D ]\t\t" + strPrint def save_algo_iteration(self, str_OSD="", newline_separator='\t\t', margin_x=25, margin_y=25, text_color=(200, 200, 200), font=cv2.FONT_HERSHEY_DUPLEX, font_scale=0.7, font_thickness=1, line_type=cv2.LINE_AA, mask_color=(255, 0, 255), img_resize_factor=0.5, save_cam_frame_before_resizing=True): if str_OSD == "": # Allow for custom OSD text, but if no text specified, print the default debug info (get_OSD_text) str_OSD = self.get_OSD_text(self.t_frame) self.t_events.append(datetime.now()) # Resize camera frame and CrazyFlie's current&target positions according to img_resize_factor curr_pos_resized = self.get_cf_curr_pos()*img_resize_factor target_pos_resized = self.get_cf_target_pos()*img_resize_factor frame_resized = cv2.resize(self.cv_cam_frame, None, fx=img_resize_factor, fy=img_resize_factor) mask_resized = cv2.resize(self.cv_filtered_HSV_mask, None, fx=img_resize_factor, fy=img_resize_factor) # Save the original camera frame to disk (for post-debugging if necessary) ###### cv2.imwrite(os.path.join(self.VIDEO_FOLDER, self.t_frame.strftime("frame_%H-%M-%S-%f.jpg")), self.cv_cam_frame if save_cam_frame_before_resizing else frame_resized) self.t_events.append(datetime.now()) # Plot OSD related to CF's current and target positions (2 circles and a connecting line) cv2.circle(frame_resized, tuple(curr_pos_resized[0:2].astype(int)), int(curr_pos_resized[2]), self.COLOR_BALL_TRACKED if self.cf_pos_tracked else self.COLOR_BALL_UNTRACKED, -1) cv2.line(frame_resized, tuple(curr_pos_resized[0:2].astype(int)), tuple(target_pos_resized[0:2].astype(int)), self.COLOR_LINE_TRACKED if self.cf_pos_tracked else self.COLOR_LINE_UNTRACKED, int(10*img_resize_factor)) cv2.circle(frame_resized, tuple(target_pos_resized[0:2].astype(int)), int(target_pos_resized[2]), self.COLOR_TARGET_TRACKED if self.cf_pos_tracked else self.COLOR_TARGET_UNTRACKED, -1) self.t_events.append(datetime.now()) # On top of that, overlay the HSV mask (so we can debug color filtering + blob detection steps) np.putmask(frame_resized, cv2.cvtColor(mask_resized, cv2.COLOR_GRAY2BGR).astype(bool), list(mask_color)) self.t_events.append(datetime.now()) # Generate the output image: upper part is the cam frame downsized according to img_resize_factor; lower part, str_OSD lines = str_OSD.split(newline_separator) self.cv_frame_out = np.zeros(((frame_resized.shape[0] + margin_y*(len(lines)+1)), frame_resized.shape[1], frame_resized.shape[2]), dtype=frame_resized.dtype) self.cv_frame_out[0:frame_resized.shape[0], :] = frame_resized self.t_events.append(datetime.now()) for cnt, l in enumerate(lines): # Add every line of text in str_OSD. Note that putText asks for bottom-left corner of text and that cnt=0 for 1st line. Therefore vertical component should be frame_resized height + OSD padding/border (0.5*margin_y) + text height (1*margin_y) cv2.putText(self.cv_frame_out, l.replace('\t', '; '), (margin_x, frame_resized.shape[0] + int(margin_y*(cnt+1.4))), font, font_scale, text_color, font_thickness, line_type) # Save the output image to disk (for post-debugging if necessary) cv2.imwrite(os.path.join(self.VIDEO_FOLDER, self.t_frame.strftime("out_%H-%M-%S-%f.jpg")), self.cv_frame_out) self.t_events.append(datetime.now()) def hover(self): try: # First, run the cv algorithm to estimate the CF's position self.detect_cf_in_camera() except: # Only way detect_cf_in_camera raises an Exception is if a camera frame couldn't be grabbed logging.exception("Couldn't grab a frame from the camera. Exiting") return datetime.now() # Need to stop now (if I wanted to stop in now+2sec and camera kept throwing exceptions, it would keep delaying the stop and never actually stop) # Now that we know where the drone currently is, send messages to control it (roll-pitch-yaw-thrust setpoints) self.control_cf() # And save the intermediate and output frames/images to disk for debugging self.save_algo_iteration() # Last, process kb input to control the experiment if not self.process_kb_input(): # Process kb input and take action if necessary (will return False when user wants to stop the experiment) self.cf_radio_connected = False # Set connected to False so next calls to this function just send thrust=0 messages return datetime.now() + timedelta(seconds=2) # For debugging purposes, it's great to have a few additional seconds of video&log after an experiment is stopped (helps see why it crashed) self.t_events.append(datetime.now()) logging.debug("DeltaT = {:5.2f}ms -> Total: {:5.2f}ms{}".format( (datetime.now() - self.t_frame).total_seconds()*1000, (self.t_frame - self.t_last_frame).total_seconds()*1000, "".join(["\t{}->{}: {}ms".format(i+1, i+2, (self.t_events[i+1]-self.t_events[i]).total_seconds()*1000) for i in range(len(self.t_events) - 1)]))) self.t_last_frame = self.t_frame # Remember the time this frame was taken so we can estimate FPS in next iteration return None # Return None to indicate that the experiment shall go on. All good. def estimate_cf_circle_depth(self): if False: t1 = datetime.now() _, contours, _ = cv2.findContours(self.cv_filtered_HSV_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) contour = contours[0] r = cv2.boundingRect(contour) c = cv2.minEnclosingCircle(contour) cv2.rectangle(self.cv_frame_out, r[0:2], (r[0] + r[2], r[1] + r[3]), (0, 255, 0), 2, cv2.LINE_AA) cv2.circle(self.cv_frame_out, tuple(int(x) for x in c[0]), int(c[1]), (0, 255, 0), 2, cv2.LINE_AA) t2 = datetime.now() print "\t\t{} ms;\t\t\tc:{}\t\tblob:{}".format((t2-t1).total_seconds()*1000, c[1], self.cf_curr_pos[2])
procs = [] # with closing(Pool(processes=2, initializer=init, initargs=(shared_arr,))) as p: # p.apply_async(producer, (v, q)) # p.apply_async(consumer, (v, q)) # p.join() procs.append(Process(target=producer, args=(v, v2, q, shared_arr))) procs.append(Process(target=consumer, args=(v, v2, q, shared_arr))) # procs.append(Process(target=end, args=(v, ))) for p in procs: p.start() for p in procs: p.join() print "Done!" exit() cap = UvcCapture(0) cap.select_best_frame_mode(60) frame = cap.get_frame_robust() cv2.imshow("", frame.bgr) key = cv2.waitKey(1) t = [] last_t = datetime.now() print "OpenCV init'ed! (consumer process says so)" while key < 0: t.append(datetime.now()) frame = cap.get_frame_robust() t.append(datetime.now()) img = frame.bgr t.append(datetime.now()) cv2.imshow("", img)