Intelligent Video Surveillance System for Real Time Effective Human Action Recognition using Deep Learning Techniques

In computer vision, recognition of human action is an important and difficult task with many practical uses, inclusive of visual recording surveillance systems and human-computer interaction. The integration of KTH and real-time datasets within a framework for human action recognition in video surveillance using deep learning. The integration of real-world data with the well-established KTH dataset notably improves the action recognition models’ generalization and robustness. The real-time dataset adds richness to the training process by capturing a variety of dynamic human actions. The system uses Convolutional Neural Networks (CNNs) to categorize behaviors in real-time video streams by extracting spatiotemporal information from video frames. The results of the experiments indicate how well this integrated dataset approach performs, and how well it can recognize a wide range of activities under different settings. This method demonstrates an overall performance of 98.7% using a real-time dataset. This breakthrough has substantially increased the dependability and efficacy of video surveillance systems insecurity applications, guaranteeing a safer environment through the effective detection and tracking of human actions in real-time scenarios.