A Comprehensive Approach for Gun, Mask and Suspicious Activity Detection using YOLOv3

Video forensics includes understanding how to examine and identify crime in the footage. The process is constantly evolving with new technology and innovations. The world of video forensics experts is growing. The realm of video forensics is simply a growing community of specialists linked with the digital video forensics sector. The State-of-the-art examinations and crimes consistently cross global and language fringes these days. With rapidly advancing technology, video has emerged as the foremost and indispensable tool in the fight against those who break the law, capturing them in the act. The computerized video forensics industry is rapidly expanding. Rapid technological advancements have made video a powerful and essential tool for law enforcement. Technology changes quickly, and innovators work with computerized images to catch criminals. Video forensics examination helps to know how accurate the input video is. In this paper, the proposed method used a soft computing technique, i.e. YOLOv3, to detect suspicious persons, the guns, or the masks by extracting frames and features from a video. In the dataset, it compares the extracted edge with images and generates output with bounding boxes for suspicious persons, the guns, or the masks. The realm of video forensics and its outcomes are also examined by this paper. When tested on different datasets, the proposed method outperforms existing techniques. For both the models, such as YOLO and customized Convolutional Neural Network (CNN), execution measurements were taken and are shown to supersede the customized CNN with its identification of guns and masks. The accuracy for YOLO design is 100% for both guns and mask detection, respectively, whereas accuracy for the customized CNN with guns and mask detection is 61.54% and 61.55%, respectively. Experimental results show that the proposed methodology outperforms the other existing methods.