Abnormal behavior detection using intelligent video surveillance

Intelligent surveillance systems have a large number of cameras installed. Abnormal vehicle or human entry at a certain location or time may potentially result in monetary loss and/or fatalities. This study develops a multi-surveillance camera intelligent surveillance system that is new, adaptable, and fast. The user may choose the number of interest zones with any polygon shape for each camera. Furthermore, the sort of abnormal item and the direction of abnormal motion for each location separately. To identify items in a video frame, the Single Shot Multi-Box Detector (SSD_MobileNet_v3) deep neural network is utilized. After that, these items are tracked using a Kernelized Correlation Filters (KCF) tracker in order to identify the direction of aberrant motion. Also, a novelty is to determine the people's motion type, i.e., running or walking, by establishing a relationship between the real human dimension and the observed distances in the video. The system's performance is evaluated on both the Authentically Distorted Surveillance Videos dataset and the newly collected dataset. An accuracy of 88.22% has been scored for event detection and F1-score of 87%. for people's motion classification. The experimental results confirm the superiority of the suggested method over the current state-of-the-art methods.