Analysis of Human-Based Suspicious Activity Using Bidirectional Long Sort Term Memory (Bi-LSTM)

Human-based Suspicious activity is that differ from regular patterns and may suggest illegal or destructive intent. This includes unexpected transactions, unauthorized access, irregular movements, or correspondence that raises concerns because it is unconventional, secretive, or potentially criminal. Public spaces are now at risk due to a sharp rise in suspicious human activities, including fighting, shooting, and fire. Market, crowd places, malls and train stations install CCTV cameras, yet data indicates their ineffectiveness without continuous video monitoring. We advise an automatic, sensible video surveillance machine for detecting and alerting on suspicious human activity. Utilizing a deep neural community, mainly the Inception V3 version, we extract key activity capabilities from video streams. These capabilities are then processed thru a Bidirectional Long Sort Term Memory (Bi-LSTM) community to establish temporal hyperlinks across frames, ensuring accurate differentiation of human movements. The system’s effectiveness is tested on records from six benchmark databases: UCF, UCF101 and UCF-Crime datasets. The strategy that was developed obtained a recognition rate of 96.80%, which is a significant improvement above the approaches that are currently considered to be state-of-the-art.