Exploration of Vehicle Target Detection Method Based on Lightweight YOLOv5 Fusion Background Modeling

Due to the explosive increase per capita in vehicle ownership China brought about by continuous development of economy and society, many negative impacts have arisen, making it necessary establish smart city system that has rapidly developing detection technology as its data acquisition system. This paper proposes a lightweight model based on an improved version YOLOv5 address problem missed false detections caused occlusion during rush hour surveillance videos. The proposed replaces BottleneckCSP structure with Ghostnet prunes network speed up inference. Additionally, Coordinate Attention Mechanism is introduced enhance network’s feature extraction improve recognition ability. Distance-IoU Non-Maximum Suppression issue omission when detecting congested targets. Lastly, combination five-frame differential method VIBE MD-SILBP operators used model’s capabilities for contours. experimental results show outperforms original terms number parameters, inference ability, accuracy applied both expanded UA-DETRAC self-built dataset. Thus, this significant industrial value intelligent traffic systems can effectively indicators monitoring scenarios.