A CNN-Based Real-Time Pose Detection System Using MediaPipe for Human Activity Monitoring

Human pose detection plays a vital role in various domains such as healthcare, sports analytics, human-computer interaction, and surveillance. This paper presents a Convolutional Neural Network (CNN)-based real-time pose detection system utilizing Google's MediaPipe framework for accurate and efficient identification of human body keypoints. The proposed system leverages deep learning techniques to predict 2D spatial positions of key joints using live webcam feeds, ensuring robust performance under diverse conditions. The CNN architecture is optimized for single-person detection and achieves a high validation accuracy of 99.3 % after extensive training. The integration of MediaPipe simplifies implementation while maintaining computational efficiency, making the system suitable for deployment in interactive applications such as fitness monitoring, gesture recognition, and fall detection. The paper also provides a comparative analysis of existing pose estimation methods and discusses the advantages of combining MediaPipe's modular design with CNN-based learning. Experimental results confirm that the proposed system delivers reliable pose predictions in real-time scenarios, with potential extensions for multi-person tracking and 3D pose estimation.