Pose Detection with OpenCV and MediaPipe

Overview

This Python script utilizes OpenCV and MediaPipe to perform real-time pose detection on a video stream. The script captures video frames, detects human poses, and displays the results with key landmarks.

Requirements

• Python 3.x
• OpenCV
• Mediapipe

Install the required libraries using:

pip install opencv-python
pip install mediapipe

Usage

1. Clone the repository or download the script.

git clone https://github.com/devDurgeshK/poseEstimationModule.git
cd poseEstimationModule

2. Run the script:

python main.py

3. Press 'q' to exit the program.

Configuration

The PoseDetector class can be configured by adjusting the initialization parameters in the script:

• staticImageMode: Set to True for static image mode.
• modelComplexity: Model complexity parameter (default: 1).
• smoothLandmarks: Enable smoothing of landmark points.
• enableSegmentation: Enable segmentation (not used in this script).
• smoothSegmentation: Enable smoothing of segmentation masks.
• minDetectionConfidence: Minimum confidence threshold for detection (default: 0.5).
• minTrackingConfidence: Minimum confidence threshold for tracking (default: 0.5).

Functionality:

1. Initialization:

   • The class is initialized with various configuration parameters related to pose detection.

2. MediaPipe Setup:

   • It initializes instances of the drawing_utils and pose modules from the mediapipe library.

3. findPose Method:

   • This method takes an image (img) as input and detects the pose using the MediaPipe Pose model.
   • If the draw parameter is True, it draws landmarks on the image.
   • Returns the modified image.

4. findPosition Method:

   • This method takes an image (img) as input and extracts the positions of pose landmarks.
   • If the draw parameter is True, it draws circles around the landmarks on the image.
   • Returns a list (lmList) containing landmark positions [id, x, y].

Note:

• The PoseDetector class encapsulates the functionality related to pose detection using the MediaPipe library.
• It provides methods for detecting poses in images and extracting pose landmark positions.
• The class is designed to be reusable and configurable with various parameters.

Contributing

Feel free to contribute to the project by submitting issues or pull requests. Your feedback and enhancements are welcomed!

Below is the functionality of the findPose and findPosition methods in the PoseDetector class:

findPose Method:

def findPose(self, img, draw=True):
    """
    Detects the pose in the given image and optionally draws the pose landmarks.

    Parameters:
    - img: The input image.
    - draw: If True, draw the pose landmarks on the image.

    Returns:
    - img: The image with or without drawn landmarks.
    """
    # Convert image to RGB format
    imgRGB = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

    # Process image with MediaPipe Pose model
    self.results = self.pose.process(imgRGB)
    
    # Draw landmarks on the image if detected
    if self.results.pose_landmarks and draw:
        self.mpDraw.draw_landmarks(img, self.results.pose_landmarks, self.mpPose.POSE_CONNECTIONS)

    return img

Functionality:

1. Image Conversion:

   • Converts the input image (img) from BGR format to RGB format, as required by the MediaPipe Pose model.

2. Pose Detection:

   • Processes the RGB image using the MediaPipe Pose model (self.pose.process()).
   • The results are stored in the self.results attribute.

3. Drawing Landmarks:

   • If the pose landmarks are detected (self.results.pose_landmarks) and the draw parameter is True, it draws the landmarks on the image using the draw_landmarks method from mediapipe.

4. Return:

   • Returns the modified image (img) with or without drawn landmarks.

findPosition Method:

def findPosition(self, img, draw=False):
    """
    Finds the positions of pose landmarks in the given image and optionally draws circles around them.

    Parameters:
    - img: The input image.
    - draw: If True, draw circles around the pose landmarks.

    Returns:
    - lmList: A list containing landmark positions [id, x, y].
    """
    lmList = []
    for id, lm in enumerate(self.results.pose_landmarks.landmark):
        # Get image dimensions
        h, w, c = img.shape
        # Convert normalized landmark coordinates to pixel values
        cx, cy = int(lm.x * w), int(lm.y * h)

        # Append landmark information to the list
        lmList.append([id, cx, cy])

        # Draw a circle around the landmark if required
        if draw:
            cv2.circle(img, (cx, cy), 7, (255, 0, 0), cv2.FILLED)

    return lmList

Functionality:

1. Landmark Iteration:

   • Iterates through the pose landmarks obtained from the MediaPipe results (self.results.pose_landmarks).

2. Coordinate Conversion:

   • Converts the normalized landmark coordinates to pixel values using the dimensions of the input image.

3. Landmark List:

   • Appends the landmark information to the lmList list in the format [id, x, y].

4. Drawing Circles:

   • If the draw parameter is True, it draws circles around the landmarks on the input image.

5. Return:

   • Returns the lmList containing landmark positions.

Note:

• findPose is responsible for detecting the overall pose in the image and drawing landmarks if required.
• findPosition extracts and returns the positions of individual pose landmarks, and optionally draws circles around them on the image.
• Both methods utilize the results obtained from the MediaPipe Pose model during the pose detection process.

Example Usage

from PoseDetector import PoseDetector
import cv2

# Create PoseDetector instance
detector = PoseDetector()

# Open video capture
cap = cv2.VideoCapture('path/to/your/video.mp4')

while True:
    success, img = cap.read()

    # Detect and draw pose
    img = detector.findPose(img)
    
    # Find and print pose landmarks
    lmList = detector.findPosition(img)
    if len(lmList) != 0:
        print(lmList)

    # Display frame with FPS
    cv2.imshow("Pose Detection", img)

    # Break the loop on 'q' key press
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

# Release the video capture and close all windows
cap.release()
cv2.destroyAllWindows()