Video Action Recognition in Noisy Environments Using Deep Learning Techniques

Accurately identifying actions in noisy environments is challenging due to video degradation and interference. Factors such as visual noise, background clutter, motion artifacts, and low-light conditions can degrade video quality, impacting the accuracy of traditional recognition systems. This project aims to tackle these challenges using advanced techniques and technologies. We leverage the Mediapipe framework, known for its robustness in multimedia processing, and integrate the high-performing video classification model Mo ViNet. Our focus is on integrating advanced noise reduction and robust action recognition methods, including spatial and temporal filtering, adaptive algorithms, deep learning, and feature extraction. By combining these techniques, we propose a comprehensive solution to enhancevideo action recognition in noisy environments, achieving higher accuracy of around 94 percent and reliability across real-world scenarios. This work contributes to advancing video action recognition systems, benefiting applications in surveillance, healthcare, interactive systems,and beyond.