Object Detection: Harmful Weapons Detection Using YOLOv4

• Published in: IEEE Symposium on Wireless Technology and Applications, ISWTA
• Main Author: Afandi W.E.I.B.W.N.; Isa N.M.
• Format: Conference paper
• Language: English
• Published: IEEE Computer Society 2021
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85125740347&doi=10.1109%2fISWTA52208.2021.9587423&partnerID=40&md5=681b238c59e48cb0e1335b42d3724fd0

Closed-circuit television (CCTV) is essential in the security industry by providing surveillance, monitoring activities, recording incidents, and storing evidence. Research and developments have been performed to ameliorate its application to meet the ever-changing security landscape. This paper presents a revolutionary method to enhance the application of CCTVs in Malaysia. The purpose of this study is to develop an Artificial Intelligence (AI) based weapons detection that helps people in identifying violent crimes that are currently happening. This study focuses on detecting harmful weapons such as handguns and knives using the custom trained object detection model that has been trained using the YOLOv4 Darknet framework. Two sets of training have been done to test the effectiveness of this system. The first training was done on a single class custom object detection model while the second was done on a multiple class custom object detection model. Based on the results obtained, the single class object detection only managed to achieve 66.67% to 77.78% accuracy on average whilst the multiple class object detection managed to achieve up to 100% accuracy on most of its input images. © 2021 IEEE