Optimizing YOLOv8 for Real-Time CCTV Surveillance: A Trade-off Between Speed and Accuracy

Abstract

Real-time video surveillance, especially CCTV systems, requires fast and accurate face detection. Object detection models with slow inference times are ineffective in real-time. This study aims to improve the inference speed of the YOLOv8 model, a preeminent object detection framework known for its accuracy and expeditiousness. The method used in this research focuses on pruning the model's architecture, particularly the P5 head section, which detects larger objects. According to Bochkovskiy's 2020 research, this modification enhances the model's performance specifically for medium and small objects depicted in CCTV footage. The standard YOLOv8 model and its modified version were compared for inference time, mean Average Precision (mAP), and model weight. This research found that the pruned YOLOv8 model demonstrated a 15.56% reduction in inference time, from 4.5 ms to 3.8 ms, accompanied by decrease in model weight. The advantages mentioned above are offset by a 1.6% decrease in mean average precision. This research contributes to the advancement of object detection technology by showcasing the architectural modifications' efficacy. These changes make the model faster and lighter, making it suitable for real-time surveillance. The accuracy trade-off is minimal. The implications of these findings are crucial for implementing efficient object detection systems in CCTV surveillance. These findings also lay the groundwork for future research to improve such systems' speed-accuracy trade-off.