Iv3-MGRUA: a novel human action recognition features extraction using Inception v3 and video behaviour prediction using modified gated recurrent units with attention mechanism model

Video analytics has become an essential tool for improving security monitoring by automating the tedious task of manually reviewing large CCTV footage. Despite advancements in this field, the accurate recognition of human actions in videos remains challenging because of the complex nature of actions, varied backgrounds, and different camera angles. To address these difficulties, we developed a novel action recognition model that integrates an attention mechanism with a modified Gated Recurrent Unit (GRU) architecture. Our approach leverages Inception v3 for feature extraction, which is combined with an attention mechanism that focuses on the most critical portions of the input sequence. This allows the model to better identify the key aspects of the video, thereby enhancing the precision of action recognition. The attention-enhanced features were further processed by the modified GRU, which utilizes an attention mechanism to categorize video behaviors more effectively, particularly for complex video sequences. To validate the effectiveness of our model, we conducted extensive tests on two well-known and challenging datasets, the Human Metabolome Database (HMDB51) and the University of Central Florida (UCF101). The results showed that our model achieved notable accuracy rates of 75.32% for HMDB51 and 96.82% for UCF101, demonstrating its capability to address the complexities of human action recognition in videos. These results highlight the potential of our approach for advancing the state-of-the-art video analytics.