Privacy-Preserving Real-Time Action Recognition on Mobile Devices Using Edge Computing and CNNs
M. S. Minu, K. Selvi, Dhanasakkaravarthi, T A Mohanaprakash, Darshan R, Abhijith Dasan, A. Krishna Kumar · 2025
The increasing use of mobile devices has created a demand for efficient, real-time video monitoring solutions optimized for mobile platforms. This study presents an innovative application leveraging Convolutional Neural Networks (CNNs) to classify human actions in videos captured on mobile devices. Trained on the UCF 101 dataset, encompassing 101 diverse action categories, the application delivers accurate recognition of a wide range of activities. By processing video data locally on the device, the solution eliminates the reliance on centralized servers, addressing latency challenges and enhancing data privacy. This approach ensures real-time performance while maintaining user confidentiality, making it ideal for sensitive applications. Key use cases include healthcare, where therapists can track patient progress during physical therapy, and sports, where the application provides real-time insights into athlete performance. The app's lightweight design and efficient resource utilization ensure compatibility with a wide range of mobile devices, supporting widespread adoption. This work advances the integration of video analysis and edge computing, delivering a scalable and privacy-preserving solution for real-time action recognition on mobile platforms.