AI-Powered Smart Physiotherapy Assistant

The project proposes an effective framework for the monitoring and classification of exercise postures using machine learning and pose estimation. The proposed system can identify the correctness of postures by analyzing the movement at the joints through the extraction of two dimensional skeletal keypoints. The proposed Neural Network model achieved the highest accuracy of 96%, showcasing strong capability in learning complex joint angle patterns and identifying subtle posture deviations. This enables users to keep good form during workouts, avoid potential injury, and enhance the effectiveness of exercises without direct guidance from professional trainers. Real-time corrective feedback is also availed to users. With this proposed method, one can enjoy home-based or gym fitness activities. Future work will expand the system to include more exercise categories and inputs from cameras at multiple angles to further improve its robustness. Integrations with real-time video processing on mobile or wearable devices will be able to expand access even more, including users training independently or remotely. Further consolidations of collaborations with physiotherapists and fitness experts will, therefore, enable clinically validated standards for posture correction, supporting applications in rehabilitation and guided therapy. In summary, this system has great potential to help individuals optimize their personal fitness training and avoid the adverse effects of harmful exercises.

[1] Z. Cao, G. Hidalgo, T. Simon, S.-E. Wei, and Y. Sheikh, “OpenPose: Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Salt Lake City, UT, Jun. 2018, pp. 7291 7299. [2] Google Research, “MediaPipe: A Framework for Building Perception Pipelines,” 2020. [Online]. Available: https://mediapipe.dev [3] S. Papoutsaki, L. Stevens, and M. Baur, “Vision-Based Rehabilitation Monitoring: Real-Time Feedback for Home Exercises,” in Proc. ACM Symp. User Interface Softw. Technol. (UIST), New Orleans, LA, Oct. 2021, pp. 45–56.  [4] W. Chan, C. Liu, and V. Jagadeesan, “Posture Correction with PoseNet: Enabling Real-Time Feedback in a Mobile Fitness Application,” in Proc. IEEE Int. Conf. on Multimedia & Expo Workshops, Shanghai, Jul. 2019, pp. 1–6.  [5] Y. Zhu, H. Wang, and J. Li, “AI in Healthcare Rehabilitation: Automated Feedback and Progress Tracking,” IEEE J. Biomed. Health Inf., vol. 26, no. 4, pp. 1875–1884, Apr. 2022.  [6] S. Chen, A. Gupta, and R. Kumar, “Pose Trainer: Correcting Yoga Postures Using Deep Learning,” in Proc. Int. Conf. on Machine Learning and Applications, Honolulu, HI, Dec. 2020, pp. 102–109.  [7] N. Tripathi, P. Singh, and M. Sharma, “Real-Time Posture Analysis for Physiotherapy,” IEEE Access, vol. 9, pp. 85 234–85 243, 2021.  [8] J. Wang, L. Zhao, and T. Brown, “Posture Monitoring for Stroke Rehabilitation with Kinect and AI,” in Proc. IEEE Int. Conf. on Rehabilitation Robotics, Toronto, ON, Jun. 2019, pp. 327–333.  [9] H. Lee and M. Park, “Posture Classification Using Convolutional Neural Networks for Exercise Error Detection,” IEEE Trans. Neural Netw. Learn. Syst., vol. 32, no. 7, pp. 2871–2882, Jul. 2021. [10] R. Kumar and S. Patel, “Smart Mirror for Posture Guidance: A   MediaPipe Powered Interactive Feedback System,” in Proc. ACM Int. Conf. on Pervasive and Ubiquitous Computing, Cancun, Mexico, Sep. 2023, pp. 617–622.