Action Recognition for Intelligent Surveillance System using LRCN with attention mechanisms

Recognizing human actions is crucial across applications like video surveillance, human-computer interaction, sports analysis, and healthcare monitoring. Despite numerous models designed for image classification in the literature, there exists a research gap in enhancing video classification models. This study presents a new approach to improve action recognition in complex surveillance environments by employing attention-based Long Short-Term Memory Recurrent Convolutional Neural Networks (LRCN). Traditional methods face challenges in handling long video sequences, occlusions, and variations in lighting conditions. LRCN, combining Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN) along with a self-attention layer, is proposed for its ability to capture both spatial and temporal information effectively. The hybrid architecture is designed to address issues such as handling long video frame sequences, occlusion, and variations in lighting, making it well-suited for real-world surveillance scenarios. The research aims to contribute to intelligent surveillance systems by improving accuracy and adaptability in recognizing actions in video frames.