Research on Object Detection of Traffic Scene Based on Deep Learning

The development of intelligent vehicle involves many key technologies. The machine vision technology based on deep learning has become one of the research hotspots because of its good performance.In this paper, aiming at the environment perception technology of smart cars, YOLOv2 deep learning algorithm is improved by combining with experimental data set, and it is applied to real-time detection of traffic scene objects.First, based on PASCAL VOC, the YOLOv2 algorithm with different network structures and loss functions was trained and tested.According to the test results, the network structure and loss function of YOLOv2 algorithm are determined.The improved YOLOv2 algorithm was trained on the images of traffic scene objects in the COCO data set, and the algorithm was tested using the actual traffic scene videos collected in the experiment to verify the detection performance of YOLOv2 algorithm on the traffic scene objects in the video.Experimental results show that YOLOv2 algorithm can obtain high detection accuracy and fast detection speed, which basically meets the technical requirements of intelligent vehicles.