Human-Pose-Estimation-System

for image

import cv2 import numpy as np import matplotlib.pyplot as plt

BODY_PARTS = { "Nose": 0, "Neck": 1, "RShoulder": 2, "RElbow": 3, "RWrist": 4, "LShoulder": 5, "LElbow": 6, "LWrist": 7, "RHip": 8, "RKnee": 9, "RAnkle": 10, "LHip": 11, "LKnee": 12, "LAnkle": 13, "REye": 14, "LEye": 15, "REar": 16, "LEar": 17, "Background": 18 }

how the body parts are connected

POSE_PAIRS = [ ["Neck", "RShoulder"], ["Neck", "LShoulder"], ["RShoulder", "RElbow"], ["RElbow", "RWrist"], ["LShoulder", "LElbow"], ["LElbow", "LWrist"], ["Neck", "RHip"], ["RHip", "RKnee"], ["RKnee", "RAnkle"], ["Neck", "LHip"], ["LHip", "LKnee"], ["LKnee", "LAnkle"], ["Neck", "Nose"], ["Nose", "REye"], ["REye", "REar"], ["Nose", "LEye"], ["LEye", "LEar"] ]

Input size

input_width = 368 input_height = 368

Load the pre-trained pose estimation model

model_path = "graph_opt.pb"
net = cv2.dnn.readNetFromTensorflow(model_path)

#for keypoints detecting confidence_threshold = 0.2

def detect_pose(frame, threshold=confidence_threshold): # dimensions frame_width = frame.shape[1] frame_height = frame.shape[0]

# Prepare the frame for input to the model
net.setInput(cv2.dnn.blobFromImage(
    frame, 1.0, (input_width, input_height), (127.5, 127.5, 127.5), swapRB=True, crop=False
))

# Perform a forward pass through the model
output = net.forward()
output = output[:, :19, :, :]  # Consider only the first 19 keypoints

# List to store the detected keypoints
keypoints = []

# Loop through all body parts to extract keypoints
for i in range(len(BODY_PARTS)):
    heat_map = output[0, i, :, :]
    _, conf, _, point = cv2.minMaxLoc(heat_map)

    x = (frame_width * point[0]) / output.shape[3]
    y = (frame_height * point[1]) / output.shape[2]

    keypoints.append((int(x), int(y)) if conf > threshold else None)

# for Drawing connections and keypoints on the frame
for pair in POSE_PAIRS:
    part_from = pair[0]
    part_to = pair[1]

    id_from = BODY_PARTS[part_from]
    id_to = BODY_PARTS[part_to]

    if keypoints[id_from] and keypoints[id_to]:
        cv2.line(frame, keypoints[id_from], keypoints[id_to], (0, 255, 0), 3)
        cv2.circle(frame, keypoints[id_from], 4, (255, 0, 0), -1)
        cv2.circle(frame, keypoints[id_to], 4, (255, 0, 0), -1)

# to Display the model's inference time
t, _ = net.getPerfProfile()
inference_time = t * 1000.0 / cv2.getTickFrequency()
cv2.putText(
    frame, f"Inference Time: {inference_time:.2f} ms", (10, 30),
    cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255), 1
)

return frame

for displaying an image using Matplotlib

def show_image(image): image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) plt.figure(figsize=(10, 10)) plt.imshow(image_rgb) plt.axis("off") plt.show()

input_source = "image.png"

try: if isinstance(input_source, str): frame = cv2.imread(input_source) if frame is None: raise FileNotFoundError(f"Image file '{input_source}' not found.") output_frame = detect_pose(frame) show_image(output_frame) cv2.imwrite("pose_output.png", output_frame) print("Pose estimation completed. Output saved as 'pose_output.png'.") else: cap = cv2.VideoCapture(input_source) print("Press 'q' to exit.") while cap.isOpened(): ret, frame = cap.read() if not ret: break output_frame = detect_pose(frame) cv2.imshow("Pose Estimation", output_frame) if cv2.waitKey(1) & 0xFF == ord('q'): break cap.release() cv2.destroyAllWindows()

except Exception as e: print(f"Error: {e}")