示例#1
0
    def __init__(self, args):
        self.folder = 'pibot_dataset/'
        if not os.path.exists(self.folder):
            os.makedirs(self.folder)
        else:
            shutil.rmtree(self.folder)
            os.makedirs(self.folder)

        # Initialise data parameters
        if args.play_data:
            self.pibot = dh.DatasetPlayer("record")
        else:
            self.pibot = PenguinPi(args.ip, args.port)
        # ckpt = ""
        self.ekf = self.init_ekf(args.calib_dir, args.ip)
        self.aruco_det = aruco.aruco_detector(self.ekf.robot,
                                              marker_length=0.07)
        if args.save_data:
            self.data = dh.DatasetWriter('record')
        else:
            self.data = None
        self.output = dh.OutputWriter('workshop_output')
        self.command = {
            'motion': [0, 0],
            'inference': False,
            'output': False,
            'save_inference': False,
            'save_image': False
        }
        self.quit = False
        self.pred_fname = ''
        self.request_recover_robot = False
        self.file_output = None
        self.ekf_on = False
        self.double_reset_comfirm = 0
        self.image_id = 0
        self.notification = 'Press ENTER to start SLAM'
        #
        self.count_down = 300
        self.start_time = time.time()
        self.control_clock = time.time()
        #
        self.detector_output = np.zeros([240, 320], dtype=np.uint8)
        self.img = np.zeros([240, 320, 3], dtype=np.uint8)
        self.aruco_img = np.zeros([240, 320, 3], dtype=np.uint8)
        if args.ckpt == "":
            self.detector = None
            self.network_vis = cv2.imread('pics/rvss_8bit/detector_splash.png')
        else:
            self.detector = Detector(args.ckpt, use_gpu=False)
            self.network_vis = np.ones((240, 320, 3)) * 100
        self.bg = pygame.image.load('pics/gui_mask.jpg')
示例#2
0
class Operate:
    def __init__(self, args):
        # Initialise data parameters
        if args.play_data:
            self.pibot = dh.DatasetPlayer(args.play_data)
        else:
            self.pibot = PenguinPi(args.ip, args.port)
        # ckpt = ""
        if args.ckpt == "":
            self.detector = None
        else:
            self.detector = Detector(args.ckpt, use_gpu=False)
        self.ekf = self.init_ekf(args.calib_dir, args.ip)
        self.aruco_det = aruco.aruco_detector(self.ekf.robot,
                                              marker_length=0.07)
        if args.save_data:
            self.data = dh.DatasetWriter('record')
        else:
            self.data = None
        self.output = dh.OutputWriter('workshop_output')
        self.command = {
            'motion': [0, 0],
            'inference': False,
            'output': False,
            'save_inference': False
        }
        self.quit = False
        self.pred_fname = ''
        self.request_recover_robot = False
        self.file_output = None
        self.ekf_on = False
        self.double_reset_comfirm = 0
        self.notification = 'Press ENTER to start SLAM'
        #
        self.count_down = 300
        self.start_time = time.time()
        self.control_clock = time.time()
        #
        self.detector_output = np.zeros([240, 320], dtype=np.uint8)
        self.img = np.zeros([240, 320, 3], dtype=np.uint8)
        self.aruco_img = np.zeros([240, 320, 3], dtype=np.uint8)
        self.network_vis = cv2.imread('pics/rvss_8bit/detector_splash.png')
        self.bg = pygame.image.load('pics/gui_mask.jpg')

    def control(self):
        if args.play_data:
            lv, rv = self.pibot.set_velocity()
        else:
            lv, rv = self.pibot.set_velocity(self.command['motion'])
        if not self.data is None:
            self.data.write_keyboard(lv, rv)
        dt = time.time() - self.control_clock
        drive_meas = measure.Drive(lv, rv, dt)
        self.control_clock = time.time()
        return drive_meas

    def take_pic(self):
        self.img = self.pibot.get_image()
        if not self.data is None:
            self.data.write_image(self.img)

    def update_slam(self, drive_meas):
        lms, self.aruco_img = self.aruco_det.detect_marker_positions(self.img)
        if self.request_recover_robot:
            is_success = self.ekf.recover_from_pause(lms)
            if is_success:
                self.notification = 'Robot pose is successfuly recovered'
                self.ekf_on = True
            else:
                self.notification = 'Recover failed, need >2 landmarks!'
                self.ekf_on = False
            self.request_recover_robot = False
        elif self.ekf_on:  # and not self.debug_flag:
            self.ekf.predict(drive_meas)
            self.ekf.add_landmarks(lms)
            self.ekf.update(lms)

    def detect_fruit(self):
        if self.command['inference'] and self.detector is not None:
            self.detector_output, self.network_vis = self.detector.detect_single_image(
                self.img)
            self.command['inference'] = False
            self.file_output = (self.detector_output, self.ekf)
            self.notification = "{} fruit type(s) detected".format(
                len(np.unique(self.detector_output)) - 1)

    def init_ekf(self, datadir, ip):
        fileK = "{}intrinsic.txt".format(datadir)
        camera_matrix = np.loadtxt(fileK, delimiter=',')
        fileD = "{}distCoeffs.txt".format(datadir)
        dist_coeffs = np.loadtxt(fileD, delimiter=',')
        fileS = "{}scale.txt".format(datadir)
        scale = np.loadtxt(fileS, delimiter=',')
        if ip == 'localhost':
            scale /= 2
        fileB = "{}baseline.txt".format(datadir)
        baseline = np.loadtxt(fileB, delimiter=',')
        robot = Robot(baseline, scale, camera_matrix, dist_coeffs)
        return EKF(robot)

    def record_data(self):
        if self.command['output']:
            self.output.write_map(self.ekf)
            self.notification = 'Map is saved'
            self.command['output'] = False
        if self.command['save_inference']:
            if self.file_output is not None:
                self.pred_fname = self.output.write_image(
                    self.file_output[0], self.file_output[1])
                self.notification = f'Prediction is saved to {operate.pred_fname}'
            else:
                self.notification = f'No prediction in buffer, save ignored'
            self.command['save_inference'] = False

    def draw(self, canvas):
        canvas.blit(self.bg, (0, 0))
        text_colour = (220, 220, 220)
        v_pad = 40
        h_pad = 20
        ekf_view = self.ekf.draw_slam_state(res=(320, 480 + v_pad),
                                            not_pause=self.ekf_on)
        canvas.blit(ekf_view, (2 * h_pad + 320, v_pad))

        robot_view = cv2.resize(self.aruco_img, (320, 240))
        self.draw_pygame_window(canvas, robot_view, position=(h_pad, v_pad))
        detector_view = cv2.resize(self.network_vis, (320, 240),
                                   cv2.INTER_NEAREST)
        self.draw_pygame_window(canvas,
                                detector_view,
                                position=(h_pad, 240 + 2 * v_pad))
        # canvas.blit(self.gui_mask, (0, 0))
        self.put_caption(canvas,
                         caption='SLAM',
                         position=(2 * h_pad + 320, v_pad))
        self.put_caption(canvas,
                         caption='Fruit Detector',
                         position=(h_pad, 240 + 2 * v_pad))
        self.put_caption(canvas, caption='PiBot Cam', position=(h_pad, v_pad))

        notifiation = TEXT_FONT.render(self.notification, False, text_colour)
        canvas.blit(notifiation, (h_pad + 10, 596))

        time_remain = self.count_down - time.time() + self.start_time
        if time_remain > 0:
            time_remain = f'Count Down: {time_remain:03.0f}s'
        elif int(time_remain) % 2 == 0:
            time_remain = "Time Is Up !!!"
        else:
            time_remain = ""
        count_down_surface = TEXT_FONT.render(time_remain, False, (50, 50, 50))
        canvas.blit(count_down_surface, (2 * h_pad + 320 + 5, 530))
        return canvas

    @staticmethod
    def draw_pygame_window(canvas, cv2_img, position):
        cv2_img = np.rot90(cv2_img)
        view = pygame.surfarray.make_surface(cv2_img)
        view = pygame.transform.flip(view, True, False)
        canvas.blit(view, position)

    @staticmethod
    def put_caption(canvas, caption, position, text_colour=(200, 200, 200)):
        caption_surface = TITLE_FONT.render(caption, False, text_colour)
        canvas.blit(caption_surface, (position[0], position[1] - 25))

    def update_keyboard(self):
        for event in pygame.event.get():
            if event.type == pygame.KEYDOWN and event.key == pygame.K_UP:
                self.command['motion'][0] = min(self.command['motion'][0] + 1,
                                                1)
            elif event.type == pygame.KEYDOWN and event.key == pygame.K_DOWN:
                self.command['motion'][0] = max(self.command['motion'][0] - 1,
                                                -1)
            elif event.type == pygame.KEYDOWN and event.key == pygame.K_LEFT:
                self.command['motion'][1] = min(self.command['motion'][1] + 1,
                                                1)
            elif event.type == pygame.KEYDOWN and event.key == pygame.K_RIGHT:
                self.command['motion'][1] = max(self.command['motion'][1] - 1,
                                                -1)
            elif event.type == pygame.KEYDOWN and event.key == pygame.K_SPACE:
                self.command['motion'] = [0, 0]
            elif event.type == pygame.KEYDOWN and event.key == pygame.K_p:
                self.command['inference'] = True
            elif event.type == pygame.KEYDOWN and event.key == pygame.K_s:
                self.command['output'] = True
            elif event.type == pygame.KEYDOWN and event.key == pygame.K_n:
                self.command['save_inference'] = True
            elif event.type == pygame.KEYDOWN and event.key == pygame.K_r:
                if self.double_reset_comfirm == 0:
                    self.notification = 'Press again to confirm CLEAR MAP'
                    self.double_reset_comfirm += 1
                elif self.double_reset_comfirm == 1:
                    self.notification = 'SLAM Map is cleared'
                    self.double_reset_comfirm = 0
                    self.ekf.reset()
            elif event.type == pygame.KEYDOWN and event.key == pygame.K_RETURN:
                n_observed_markers = len(self.ekf.taglist)
                if n_observed_markers == 0:
                    if not self.ekf_on:
                        self.notification = 'SLAM is running'
                        self.ekf_on = True
                    else:
                        self.notification = '> 2 landmarks is required for pausing'
                elif n_observed_markers < 3:
                    self.notification = '> 2 landmarks is required for pausing'
                else:
                    if not self.ekf_on:
                        self.request_recover_robot = True
                    self.ekf_on = not self.ekf_on
                    if self.ekf_on:
                        self.notification = 'SLAM is running'
                    else:
                        self.notification = 'SLAM is paused'
            elif event.type == pygame.QUIT:
                self.quit = True
            elif event.type == pygame.KEYDOWN and event.key == pygame.K_ESCAPE:
                self.quit = True
        if self.quit:
            pygame.quit()
            sys.exit()