Deep Learning-based Target Recognition in Complicated Traffic Scenes

The ongoing evolution of urban transportation has complicated the analysis of traffic footage. With the fast growth of deep learning theory, good results in object detection in basic settings have been obtained. However, in complicated settings, mixed target recognition of people and electric vehicles running red lights is a critical and tough problem for traffic management. The difficulty of object recognition in complicated traffic scenarios is the topic of this paper. The object detection model is improved by effectively integrating the Retinex image enhancement network with bilateral filtering Gaussian pyramid transform and YOLOV4 network to solve the problem of poor object detection performance caused by dim night light and poor multi object detection performance in complex environments. This model fits the application requirements of actual complicated traffic scenarios through experimental comparison and testing, both in terms of detection effectiveness and detection speed.