Subduing Crime and Threat in Real-Time by Detecting Weapons Using Yolov8

With the rapid development of large cities and towns, crime rate graphs are also increasing steadily. The crime rate has increased along with the urban and rural areas' rapid development. Closed Circuit Television (CCTV) has been deployed in virtually every location, including airports, shopping centers, apartments, etc., to protect public safety. In other words, it has been installed everywhere people congregate. Our study focuses on detecting firearms, knives, and other sharp objects in CCTV footage in close to real-time while assessing the threat level based on the subject's arm position. You Only Look Once(YOLO) version 8 analyses CCTV footage to discover the presence of a knife or gun or other sharp objects. The evaluation of the subject's arm posture is done if a weapon is found. The threat level is determined based on how the arm holding the weapon is held. Since there are no available standard datasets, we created our own custom dataset using information from websites and Github projects. Among the tools used for object detection, annotation, and pose detection are YOLOV8, Roboflow, and Mediapipe. When performing object detection, precision and recall matter more than accuracy, hence the full set of methods was evaluated in terms of these. Nearly 6000 images were used to train the model. The model achieved an accuracy of 93%.