def run():
# Initialize keypoints extractor
kp_extractor = KeypointsExtractor()
base_path = "/home/kevin/projects/initial-pose-data/images/pos-negs"
for exercise_name in os.listdir(base_path):
exercise_dir = os.path.join(base_path, exercise_name)
for posneg in os.listdir(exercise_dir):
posneg_dir = os.path.join(exercise_dir, posneg)
for file in os.listdir(posneg_dir):
file_path = os.path.join(posneg_dir, file)
image = cv2.imread(file_path)
try:
list_of_pose_and_id, _ = kp_extractor.get_keypoints_and_id_from_img(
image)
except Exception as e:
# cv2.imshow("image yg error", image)
# cv2.waitKey(0)
print(e)
continue
keypoints = list_of_pose_and_id[0]['Keypoints']
if validate_keypoints(keypoints):
keypoints = keypoints.tolist()
is_starting_pose = True if posneg == "pos" else False
insert_starting_pose_binary_to_db(keypoints, exercise_name,
is_starting_pose)
else:
cv2.imshow("image yg ga ke validate", image)
cv2.waitKey(0)
def run():
# Initialize paths
base_path = "/home/kevin/projects/initial-pose-data/train_data"
save_path = "/home/kevin/projects/exercise_pose_evaluation_machine/models/initial_pose_model/initial_pose_model.h5"
# Get dataset folders
dirs = os.listdir(base_path)
# Initialize keypoints extractor
kp_extractor = KeypointsExtractor()
# Logger
kp_extraction_logger = create_logger()
kp_extraction_logger.write_header()
min_max_scaler = MinMaxScaler(feature_range=(0, 1))
for class_name in dirs:
class_dir = os.listdir(base_path + '/' + class_name)
for file_name in class_dir:
file_path = f'{base_path}/{class_name}/{file_name}'
image = cv2.imread(file_path)
# Get image dimensions for logging
height, width, channels = image.shape
# record the time to extract keypoints
kp_extract_start = time.perf_counter()
list_of_pose_and_id = kp_extractor.get_keypoints_and_id_from_img(
image)
kp_extract_end = time.perf_counter()
kp_extraction_log_entry = {
"image_res": f"{width}x{height}",
"time": kp_extract_end - kp_extract_start,
"exercise_type": class_name
}
kp_extraction_logger.write_row(kp_extraction_log_entry)
keypoints = list_of_pose_and_id[0]['Keypoints']
if validate_keypoints(keypoints):
normalized_keypoints = min_max_scaler.fit_transform(keypoints)
normalized_keypoints = normalized_keypoints.tolist()
# print(normalized_keypoints)
# print(type(normalized_keypoints))
insert_initial_pose_to_db(normalized_keypoints, class_name)
def run():
# Initialize keypoints extractor
kp_extractor = KeypointsExtractor()
base_path = "/home/kevin/projects/initial-pose-data/images/v2/pos-negs"
for exercise_name in os.listdir(base_path):
exercise_dir = os.path.join(base_path, exercise_name)
for posneg in os.listdir(exercise_dir):
posneg_dir = os.path.join(exercise_dir, posneg)
for file in os.listdir(posneg_dir):
file_path = os.path.join(posneg_dir, file)
image = cv2.imread(file_path)
list_of_pose_and_id, _ = kp_extractor.get_keypoints_and_id_from_img(image)
keypoints = list_of_pose_and_id[0]['Keypoints']
if validate_keypoints(keypoints):
keypoints = keypoints.tolist()
is_starting_pose = True if posneg == "pos" else False
insert_starting_pose_binary_to_db(keypoints, exercise_name, is_starting_pose)
def test():
pose_detector = PoseDetector("squat")
kp_extractor = KeypointsExtractor()
file_path = '/home/kevin/projects/initial-pose-data/images/v1/pos-negs/squat/pos/mirrored-VID_20200928_134747.mp4_186.jpg'
# Opening OpenCV stream
stream = cv2.VideoCapture(file_path)
while True:
try:
ret, imageToProcess = stream.read()
except Exception as e:
# Break at end of frame
break
list_of_keypoints = kp_extractor.get_keypoints_and_id_from_img(imageToProcess)
x = list_of_keypoints[0]
keypoints = np.array(x).flatten()
prediction = pose_detector.predict(np.array([keypoints]))
print(prediction)
def __init__(self):
""" Virtually private constructor. """
if PoseMachine.__instance != None:
raise Exception("This class is a singleton!")
else:
self.target_id = ""
self.exercise_type = ""
self.pose_predictions = []
self.kp_extractor = KeypointsExtractor()
self.init_pose_detector = InitialPoseDetector()
self.right_hand_up_detector = RightHandUpDetector()
PoseMachine.__instance = self
from sklearn.preprocessing import OneHotEncoder
from sklearn.model_selection import train_test_split
from keypoints_extractor import KeypointsExtractor
def predict_initial_pose(keypoints):
# Base paths
save_path = "/home/kevin/projects/exercise_pose_evaluation_machine/models/initial_pose_model/initial_pose_model.h5"
# Prepare keypoints to feed into network
keypoints = np.array(keypoints).flatten().astype(np.float32)
# Load model
model = load_model(save_path)
# Summarize model
model.summary()
# Get prediction
res_lookup = ['plank', 'sit-up', 'dumbell-curl', 'push-up']
return res_lookup[np.argmax(model.predict(np.array([keypoints])))]
if __name__ == '__main__':
img_test_path = "/home/kevin/projects/Exercise Starting Pose/Pushups/pushupstart9.jpg"
image = cv2.imread(img_test_path)
kp_extractor = KeypointsExtractor()
list_of_pose_and_id = kp_extractor.get_keypoints_and_id_from_img(image)
keypoints = list_of_pose_and_id[0]['Keypoints']
print("Result is " + predict_initial_pose(keypoints))
"push-up": "/home/kevin/projects/dataset/push-up-t.mp4", # OK
"sit-up": "/home/kevin/projects/dataset/sit-up-t.mp4", # OK
"squat": "/home/kevin/projects/dataset/squat-k1.mp4", # OK
"plank": "/home/kevin/projects/dataset/plank-k1.mp4", # OK
# "test": "/home/kevin/Videos/VID_20210116_152545.mp4"
}
if __name__ == '__main__':
try:
# Base paths
exercise = "plank"
base_path = VIDEO_PATHS[exercise]
# base_path = "/home/kevin/projects/dataset/squat-obscured.mp4"
# base_path = "/home/kevin/projects/right-hand-up-to-exercise/squat.mp4"
kp_extractor = KeypointsExtractor()
# Opening OpenCV stream
stream = cv2.VideoCapture(base_path)
fps = stream.get(cv2.CAP_PROP_FPS)
# Define flags
init_pose_detected = False
target_id = -1
# Define detectors
right_hand_up_detector = RightHandUpDetector()
initial_pose_detector = InitialPoseDetector()
# Define x_min, y_min, x_max, y_max
x_min = -1
normalized_keypoints = [x.tolist() for x in keypoints]
insert_right_hand_up_pose_to_db(normalized_keypoints, label)
if __name__ == "__main__":
from multiprocessing import Process
# Initialize paths
# TODO: Setup the train data in 2 folders true and false folders
base_path = "/home/kevin/projects/right_hand_up_pose_data/training_data"
# Get dataset folders
dirs = os.listdir(base_path)
# Initialize keypoints extractor
kp_extractor = KeypointsExtractor()
min_max_scaler = MinMaxScaler(feature_range=(0, 1))
# Label should be either "true" / "false" (Which are the folder names)
for label in dirs:
label_dir = os.listdir(base_path + '/' + label)
run(base_path, label, label_dir, kp_extractor, min_max_scaler)
# label_dir_split = chunks(label_dir, int(len(label_dir) / 3))
# THREADS = []
# for label_dir in label_dir_split:
# thread = Process(target=run, args=(base_path, label, label_dir,))
# thread.start()
# THREADS.append(thread)
# for t in THREADS:
return False
return True
if __name__ == '__main__':
# Keypoint list for each exercise
selected_keypoints = [
NECK, R_SHOULDER, R_ELBOW, R_WRIST, L_SHOULDER, L_ELBOW, L_WRIST,
MID_HIP, R_HIP, R_KNEE, R_ANKLE, L_HIP, L_KNEE, L_ANKLE
]
# Initialize dataset path
dataset_path = '/home/kevin/projects/dataset-theo/to-insert-to-db'
# Define keypoints
kp_extractor = KeypointsExtractor()
# Get all keypoints from dataset
# and insert into database
dirs = os.listdir(dataset_path)
for foldername in dirs:
# Get all folder list
folder = dataset_path + '/' + foldername
files = os.listdir(folder)
# Repeat for every files
for filename in files:
try:
keypoints = []
class_type = 0
print("Processing file: " + filename)
L_EYE = 16
R_EAR = 17
L_EAR = 18
L_BIG_TOE = 19
L_SMALL_TOE = 20
L_HEEL = 21
R_BIG_TOE = 22
R_SMALL_TOE = 23
R_HEEL = 24
selected_keypoints = [
NECK, R_SHOULDER, R_ELBOW, R_WRIST, L_SHOULDER, L_ELBOW, L_WRIST,
MID_HIP, R_HIP, R_KNEE, R_ANKLE, L_HIP, L_KNEE, L_ANKLE
]
# Get reps keypoints
kp_extractor = KeypointsExtractor()
all_exercise_reps, all_exercise_x_low, all_exercise_y_low = kp_extractor.scan_video_without_normalize(
base_path, selected_keypoints)
scaler = make_min_max_scaler(all_exercise_reps, min(all_exercise_x_low),
min(all_exercise_y_low))
normalized_reps = normalize_keypoints_from_external_scaler(
all_exercise_reps, scaler)
reshaped_normalized_reps = [
np.array(frames).flatten() for frames in normalized_reps
]
# Print results
print("Result is " + predict_sequence(reshaped_normalized_reps, "sit-up"))
# JAPHNE MADAFAKIN DID IT AGAIN!
# JAPHNE MADAFAKIN DID IT AGAIN!
# JAPHNE MADAFAKIN DID IT AGAIN!