Edge Computing-Based Crime Scene Object Detection from Surveillance Video Using Deep Learning Algorithms

Crime scene artifacts include a handgun, revolver, or gun are utilised in most criminal crimes. A human must intervene in the surveillance and control system to discover such illegal activity. The purpose of this study is to create an automated, real-time object detection system for use with surveillance cameras at crime scenes. The intention is to warn security personnel of potentially dangerous situations. The proposed Fully connected Convolutional neural networks (CNNs) combined with Edge Computing so it is called hybrid system, and faster region-convolutional neural networks (faster R-CNNs), and regression are all examples of deep learning algorithms. Among the most promising approaches for detecting objects now in use are linear-based detectors YoloV7. Real-time detection requires a faster detection speed than region-based algorithms can provide. On the other hand, real-time detection may be achieved using regression-linear-based detectors such as YoloV7. In this research, YoloV7 was used to identify crime objects quickly. Crime object, and then Fully Connected CNN was used to improve detection accuracy. The system is trained and tested using a crime object dataset consisting of 5,660 images of weapons and anomaly action that are both publicly and privately accessible, with an 80-20 split, and then assessed using live security camera footage. In addition to its 92% accuracy, 95% precision, 88% recall, and an F1 score of 88.5%. The resultant findings validate the use of the suggested hybrid system for real-time monitoring by indicating the presence of a suspect carrier in an area under observation.