def preprocess_output(self, eye_boxes, face_image): """ Scales the eye bounding boxes from the 0..1 range to the size of the face image. Returns a tuple of cropped left and right eye images. """ if face_image is not None and face_image.size > 0 and eye_boxes and eye_boxes[ 0] and eye_boxes[1]: left_eye_box = helpers.fit(eye_boxes[0], face_image) left_eye = helpers.crop(face_image, left_eye_box) right_eye_box = helpers.fit(eye_boxes[1], face_image) right_eye = helpers.crop(face_image, right_eye_box) return left_eye, right_eye else: return None, None
def captureHistogram(source=None): source = 0 if source is None else source if source is not None and str(type(source)) != "<class 'int'>": raise ValueError("source: integer value expected") cap = cv2.VideoCapture(source) boxSize = 100 centerX = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH) / 2) centerY = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT) / 3) padding = boxSize + 10 while True: _, frame = cap.read() frame = cv2.flip(frame, 1) box1 = (centerY + padding * 0, centerX + padding * 0, boxSize, boxSize) box2 = (centerY + padding * 0, centerX + padding * 1, boxSize, boxSize) box3 = (centerY + padding * 1, centerX + padding * 0, boxSize, boxSize) box4 = (centerY + padding * 1, centerX + padding * 1, boxSize, boxSize) box5 = (centerY + padding * 2, centerX + padding * 0, boxSize, boxSize) box6 = (centerY + padding * 2, centerX + padding * 1, boxSize, boxSize) helpers.drawBoxes(frame, helpers.PURPLE, [box1, box2, box3, box4]) boxOne = helpers.crop(frame, box1) boxTwo = helpers.crop(frame, box2) boxThree = helpers.crop(frame, box3) boxFour = helpers.crop(frame, box4) boxFourBlurred = helpers.blur(boxFour, 15) helpers.copyTo(frame, box5, boxFour) helpers.copyTo(frame, box6, boxFourBlurred) cv2.imshow("Video Feed", frame) key = cv2.waitKey(10) if key == 97: objectColor = finalHistImage cv2.destroyAllWindows() break if key == 27: cv2.destroyAllWindows() cap.release() exit() hsvObjectColor = cv2.cvtColor(objectColor, cv2.COLOR_BGR2HSV) objectHist = cv2.calcHist([hsvObjectColor], [0, 1], None, [12, 15], [0, 180, 0, 256]) cv2.normalize(objectHist, objectHist, 0, 255, cv2.NORM_MINMAX) return objectHist
def recognize_gesture(image, handDetector, signDetector, label_encoder): #find hand #please play with the hyperparameters to find a balance between accuracy and runtime detected_box = sliding_window(handDetector, image, 50, .9) #crop image around hand croppedImage = helpers.crop(image, int(detected_box[0]), int(detected_box[2]), int(detected_box[1]), int(detected_box[3])) #get HOG of hand hogvec = helpers.convertToGrayToHOG(croppedImage) #use our trained model to get the probability of each class prediction_probs = signDetector.predict_proba([hogvec])[0] #get predicted class based on predicted probabilities prediction = label_encoder.inverse_transform( prediction_probs.tolist().index(max(prediction_probs))) return detected_box, prediction_probs, prediction #handDetector = helpers.loadClassifier('handDetector.pkl') # #signDetector = helpers.loadClassifier('signDetector.pkl') # #label_encoder = LabelEncoder().fit(['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'K', 'L', 'M', 'N', # 'O', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y']) # #image = helpers.io.imread('Dataset/user_3/S2.jpg') # #loc, probs, pred = recognize_gesture(image, handDetector, signDetector, label_encoder) # #print(probs)
def captureHistogram(source=None): source = 0 if source is None else source if source is not None and str(type(source)) != "<class 'int'>": raise ValueError("source: integer value expected") cap = cv2.VideoCapture(source) boxSize = 30 centerX = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH) / 2) centerY = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT) / 2) while True: _, frame = cap.read() frame = cv2.flip(frame, 1) cv2.putText(frame, "Place region of the hand inside the boxes and press `A`", (5, 50), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 255), 1, cv2.LINE_AA) box1 = (centerX + 0, centerY + 0, boxSize, boxSize) box2 = (centerX + 50, centerY + 0, boxSize, boxSize) box3 = (centerX + 0, centerY + 50, boxSize, boxSize) box4 = (centerX + 50, centerY + 50, boxSize, boxSize) helpers.drawBoxes(frame, helpers.PURPLE, [box1, box2, box3, box4]) boxOne = helpers.crop(frame, box1) boxTwo = helpers.crop(frame, box2) boxThree = helpers.crop(frame, box3) boxFour = helpers.crop(frame, box4) finalHistImage = np.hstack((boxOne, boxTwo, boxThree, boxFour)) cv2.imshow("Video Feed", frame) key = cv2.waitKey(10) if key == 97: objectColor = finalHistImage cv2.destroyAllWindows() break if key == 27: cv2.destroyAllWindows() cap.release() exit() hsvObjectColor = cv2.cvtColor(objectColor, cv2.COLOR_BGR2HSV) objectHist = cv2.calcHist([hsvObjectColor], [0, 1], None, [12, 15], [0, 180, 0, 256]) cv2.normalize(objectHist, objectHist, 0, 255, cv2.NORM_MINMAX) return objectHist
def preprocess_output(self, face_box, frame): """ Scales the face bounding box from the range 0..1 to the size of the frame and extracts the face image. Returns a tuple containing the face image and the scaled bounding box. """ if face_box is None or frame is None or frame.size < 1: return None, None else: face_box = helpers.fit(face_box, frame) face_img = helpers.crop(frame, face_box) return face_img, face_box
def buildhandnothand_lis(frame, imgset): poslis = [] neglis = [] for nameimg in frame.image: tupl = frame[frame['image'] == nameimg].values[0] x_tl = tupl[1] y_tl = tupl[2] side = tupl[5] conf = 0 dic = [0, 0] arg1 = [x_tl, y_tl, conf, side, side] poslis.append( helpers.convertToGrayToHOG( helpers.crop(imgset['Dataset/' + nameimg], x_tl, x_tl + side, y_tl, y_tl + side))) while dic[0] <= 1 or dic[1] < 1: x = random.randint(0, 320 - side) y = random.randint(0, 240 - side) crp = helpers.crop(imgset['Dataset/' + nameimg], x, x + side, y, y + side) hogv = helpers.convertToGrayToHOG(crp) arg2 = [x, y, conf, side, side] z = overlapping_area(arg1, arg2) if dic[0] <= 1 and z <= 0.5: neglis.append(hogv) dic[0] += 1 if dic[0] == 1: break label_1 = [1 for i in range(0, len(poslis))] label_0 = [0 for i in range(0, len(neglis))] label_1.extend(label_0) poslis.extend(neglis) return poslis, label_1
def get_data(user_list, img_dict, data_directory): X = [] Y = [] for user in user_list: user_images = glob.glob(data_directory + user + '/*.jpg') boundingbox_df = pd.read_csv(data_directory + user + '/' + user + '_loc.csv') for rows in boundingbox_df.iterrows(): cropped_img = helpers.crop( img_dict[data_directory + rows[1]['image']], rows[1]['top_left_x'], rows[1]['bottom_right_x'], rows[1]['top_left_y'], rows[1]['bottom_right_y']) hogvector = helpers.convertToGrayToHOG(cropped_img) X.append(hogvector.tolist()) Y.append(rows[1]['image'].split('/')[1][0]) return X, Y
def cacheSteps(imgset, frame, step_x, step_y): # print "Cache-ing steps" list_dic_of_hogs = [] dic = {} i = 0 for img in frame.image: tupl = frame[frame['image'] == img].values[0] x_tl = tupl[1] y_tl = tupl[2] side = tupl[5] conf = 0 i += 1 if i % 10 == 0: print(i, " images cached") image = imgset['Dataset/' + img] for x in range(0, 320 - side, step_x): for y in range(0, 240 - side, step_y): dic[str(img + str(x) + str(y))] = helpers.convertToGrayToHOG( helpers.crop(image, x, x + side, y, y + side)) return dic
gazeEstimator.feed_input(left_eye, right_eye, head_pose_angles) # get a new head pose if available head_pose_consumed, head_pose_angles = headPoseEstimator.consume_output( wait_needed) if head_pose_consumed: q[head_poses_produced][3] = head_pose_angles head_poses_produced += 1 # get a new pair of eye images if available eyes_consumed, eye_boxes = eyeDetector.consume_output(wait_needed) if eyes_consumed: frame, face_box, _, _ = q[ eyes_produced] # the corresponding frame and the face bounding box for this detection eyes = eyeDetector.preprocess_output(eye_boxes, face_image=helpers.crop( frame, face_box)) q[eyes_produced][2] = eyes eyes_produced += 1 # get a new face and the bounding box if available face_consumed, face_box = faceDetector.consume_output( confidence=args.confidence, wait=wait_needed) if face_consumed: face_img, face_box = faceDetector.preprocess_output( face_box, frame=q[faces_produced][0]) q[faces_produced][1] = face_box faces_produced += 1 eyeDetector.feed_input(face_img) headPoseEstimator.feed_input(face_img) frame = feed.read_next() # get the next frame from the input feed
def crop(self): size_list = [(128, 128), (256, 256), (512, 512), (1024, 1024), (128, 0), (256, 0), (512, 0), (1024, 0)] for size in size_list: helpers.crop(self.path, size)
def crop(self): size_list = [(128, 128), (256, 256), (512, 512), (1024, 1024), (128, 0), (256, 0), (512, 0), (1024, 0)] for size in size_list: helpers.crop(self.path, size)