Internship Project under AICTE's Virtual Internship Program focused on applying deep learning and machine learning concepts for human pose detection from images and video frames.
This project was completed as part of the AICTE Machine Learning Internship, where the primary objective was to build an ML model capable of detecting and estimating human body keypoints in real-time.
Human pose estimation refers to the process of detecting human joints (keypoints) such as elbows, knees, shoulders, etc., from images or video — which has critical applications in healthcare, sports analysis, surveillance, and human-computer interaction.
As an intern, I was actively involved in:
- Understanding the theoretical background of pose estimation
- Implementing models using OpenPose and MediaPipe
- Preprocessing image datasets for training and inference
- Annotating keypoints and evaluating model accuracy
- Deploying the model for real-time pose detection using webcam input
- Python 3
- OpenCV
- TensorFlow / PyTorch
- MediaPipe
- NumPy, Matplotlib
- Jupyter Notebook
- Real-time pose detection from webcam/video
- Visualization of 18 keypoints with skeleton overlay
- Integration with MediaPipe Pose and OpenPose for accuracy comparison
- Heatmap generation for joint probability analysis
- Option to track multiple persons
- Achieved real-time performance of ~20 FPS on CPU using MediaPipe
- Successfully detected keypoints with ~85% accuracy on sample datasets
- Visualized joint connections with labeled skeleton overlays
- Developed a strong understanding of CNNs and keypoint detection models
- Gained practical skills in image processing and ML model deployment
- Improved ability to visualize and evaluate pose estimation outputs
- Hands-on experience with real-time video processing using OpenCV
Special thanks to AICTE and project mentors for their valuable guidance and structured learning framework during the internship.
- Name: Bipasha Acharjee
- LinkedIn: View my profile
⭐ If this repository helped you, give it a star and feel free to fork for collaboration!