Dynamic Suspicious Activity Detection using SSA-Optimized Deep CNN in Surveillance Videos

Ensuring public safety and security is of paramount importance in today's world, and the need for effective human suspicious activity detection in surveillance videos has never been greater. Detecting unusual or potentially harmful behaviors automatically provides a means for timely intervention, reducing response time, and assisting law enforcement agencies in maintaining order. This research introduces a novel solution that leverages the SSAAlgorithm to enhance the capabilities ofa deep CNN classifier, enabling more accurate and reliable identification of suspicious activities. The research delves into person detection and tracking, facilitating the monitoring of individuals within the video footage. Image skeletonization techniques are applied to outline the core structures of objects and shapes within the images. The computation of statistical features, grid features, and the extraction of Histogram of Oriented Optical Flow (HOOF) features. These extracted features serve as vital inputs for the deep CNN classifier, which excels in identifying abnormal activities effectively. The innovative S SA approach enables the system to adapt dynamically to changing scenarios and emerging threats which is inspired by the food-searching traits of sparrows. The research demonstrates superior results, as measured by accuracy 93.34%, sensitivity 87.18%, and specificity 95.00% for TP 90.