Enhanced YOLOv4 for Facilitating Public Safety Management amidst Protests and Riots
Kanagasabai Thiruthanigesan, Ruwan Dharshana Nawarathna, Roshan Ragel · 2023
Public safety management during protests is a critical issue requiring effective and efficient measures to minimise harm to law enforcement personnel and protesters. Our research introduces a faster variant of the You Only Look Once version 4 (YOLOv4) algorithm, specifically designed to identify objects frequently employed as weapons or for starting fires in public demonstrations and riots. We also developed a new dataset containing knives, swords, axes, stones, sticks, and objects commonly associated with arson. The proposed Accelerated YOLOv4 algorithm was trained and tested using the developed dataset. Its performance was compared to the standard YOLOv4 algorithm. The performance evaluation of the detection models showed that the Accelerated YOLOv4 model achieved higher precision scores than the YOLOv4 model for most objects. The mean average precision (mAP) of the Accelerated YOLOv4 model was 86%, which is higher than the mAP of the YOLOv4 model, which was 72%. These results indicate that the proposed Accelerated YOLOv4 algorithm can improve the effective management of public safety during protests.