import cv2 as cv import mediapipe as mp from camera import frame_capture import csv mp_hands = mp.solutions.hands # type: ignore[attr-defined] mp_draw = mp.solutions.drawing_utils # type: ignore def initialize_csv(): try: with open("train_data.csv", 'r'): pass except: with open("train_data.csv", 'a') as train_init: writer = csv.writer(train_init, delimiter=',') header=[] for i in range(len(mp_hands.HandLandmark)): header.append(mp_hands.HandLandmark(i).name + "_x") header.append(mp_hands.HandLandmark(i).name + "_y") header.append(mp_hands.HandLandmark(i).name + "_z") header.append("label") writer.writerow(header) initialize_csv() hand_detector = mp_hands.Hands( static_image_mode=False, max_num_hands=1, min_detection_confidence=0.2, min_tracking_confidence=0.5 ) def clean_data(hand_num, hand_landmarks): distinct_landmarks = [(mp_hands.HandLandmark(idx).name, landmark) for idx, landmark in enumerate(hand_landmarks)] #return list of tuples return distinct_landmarks def clean_coordinates_for_csv(list_landmarks): wrist_coordinates = list_landmarks[0][1] coordinates = [] for _, x_y_z in list_landmarks: coordinates.append(x_y_z.x - wrist_coordinates.x) coordinates.append(x_y_z.y - wrist_coordinates.y) coordinates.append(x_y_z.z - wrist_coordinates.z) #note: i subtract each landmark coordinate from the wrist to help the classifier with translation invariance return coordinates def run(): frames = frame_capture() for frame_dict in frames: rgb = frame_dict['rgb'] bgr = frame_dict['bgr'] processed_hand = hand_detector.process(rgb) #process the rgb version of the frame if processed_hand.multi_hand_landmarks: # if non zero list for hand_num, hand_landmarks in enumerate(processed_hand.multi_hand_landmarks): #per hand mp_draw.draw_landmarks( bgr, #draw on original bgr frame of opencv hand_landmarks, mp_hands.HAND_CONNECTIONS ) landmarks = clean_data(hand_num, hand_landmarks.landmark) digit_pressed = cv.waitKey(1) if digit_pressed in [ord('1'), ord('2'), ord('3'), ord('4'), ord('5')]: coordinates = clean_coordinates_for_csv(landmarks) coordinates.append(digit_pressed - 48) #append the label (-48 to get num not ascii) with open("train_data.csv", 'a', newline='') as train: writer = csv.writer(train, delimiter=',') writer.writerow(coordinates) print(coordinates) # print(f"Hand_{hand_num}: Landmarks: {landmarks}") cv.imshow("frame", bgr) if cv.waitKey(1) == ord('q'): break if __name__ == '__main__': run()