Loitering Based Human Crime Detection in Video Surveillance using Beluga Whale Adam Dingo Optimizer and Deep Convolutional Neural Network

Abstract

Video Surveillance (VS) systems play a crucial role in maintaining security in public spaces, commercial establishments and residential areas. Detecting and preventing human-related crimes within the footage captured by these systems is a challenging task. Traditionally, VS systems rely on basic motion detection, which often leads to false alarms and inefficient use of resources. Loitering, a behavior frequently associated with criminal activities, requires more nuanced detection to reduce false positives and improve response times. Accurate tracking of individuals, especially in crowded environments, is another challenge. The chief objective of this research is to address these challenges by introducing an innovative Loitering-based Human Crime Detection (LHCD) module in VS. This module combines enhanced euclidean based Deep Simple Online Real-time Tracking (DSORT) with the Segmentation Quality Assessment (SQA) algorithm to accurately assess human travel distances. Also, this research integrates the Beluga Whale Adam Dingo Optimizer (BWADO) and a Deep Convolutional Neural Network (DCNN) to boost the precision and efficiency of Human Crime Detection (HCD) within loitering areas. The introduced approach demonstrates the effectiveness of introduced module, which reduces false alarms and enhances response times in VS. Outcomes demonstrate that the introduced approach outperforms existing approaches in various performance measures like accuracy (99.76%), F1-score (99.89%), recall (98.59%), precision (98.9%) and processing time (1.78s) demonstrating its superior effectiveness and potential for advancements in the field