Research On Night Vehicle Detection Method Based On Deep Learning

Vehicle detection technology has become a hot topic in the field of computer vis ion in recent years.At present,algorithms related to vehicle detection have made rem arkable achievements in daytime environment,but the problem of target vehicle dete ction in nighttime environment has not been effectively solved.Aiming at various roa d scenes in the nighttime environment,this paper analyzes various difficult problems existing in the current vehicle detection task,focuses on the research on the enhance ment algorithm of nighttime images and the target detection algorithm based on impr oved YOLOv5,finally designs a deep learning algorithm suitable for nighttime vehic le detection.The main contents and achievements of this paper are as follows:(1)Research on image enhancement algorithm in night environment.The featur e information of night vehicles is easy to be hidden because the road background is t oo dark,so in order to reduce the difficulty of feature extraction,this paper selects m ulti-scale Retinex algorithm with color recovery for image enhancement of night vehi cles.The nighttime images processed by this algorithm are enhanced in brightness an d contrast,effectively distinguish the color difference between the road background a nd the target vehicle,retain more effective information,and reduce the difficulty of f eature extraction.(2)A night vehicle detection algorithm based on improved YOLOv5 is designed.Vehicle targets in road scenes often have the characteristics of scale diversity.This p aper is based on YOLOv5 algorithm,the method is to integrate AFF multi-scale featu re fusion module into the network structure to enhance the ability of feature extractio n at different scales.Through experimental verification,the optimization algorithm i mproves the m AP value by 5.5 percentage points,and compared with the original Y OLOv5 algorithm model,the average accuracy,precision and recall of the algorithm have been greatly improved.(3)In order to improve the detection algorithm’s attention to target vehicles,this paper adds coordinate attention mechanism to the improved YOLOv5 network.It av oids missing inspection in the scene of dense traffic at night.Through experimental a nalysis,the method greatly improves the performance of model detection,and the ac curacy rate is increased by 5.92 percentage points,which is more suitable for vehicle detection tasks in the nighttime environment.