Research On Pedestrian And Vehicle Detection Algorithm Based On Deep Learning

With the rapid development of urbanization and the aggregation of population,road congestion is becoming more and more serious,which brings about problems such as too long commute time and frequent traffic accidents.Pedestrian and vehicle detection algorithms based on deep learning can greatly liberate labor and reduce urban road management pressure.It plays a huge role in improving road congestion and reducing violations and traffic accidents effectively,and has great research and application value.In order to complete the task of detecting pedestrians and vehicles in actual road scenes,this paper is based on the YOLOv3 target detection algorithm,combined with the realistic requirements of pedestrian vehicle detection,and proposes a pedestrian vehicle detection algorithm based on multi-scale feature fusion and attention mechanism.The main research contents are:(1)Aiming at the problems of YOLOv3 that the detection accuracy of small and medium-scale targets is not high,the positioning accuracy is insufficient,and the detection speed is not fast enough,a pedestrian and vehicle detection algorithm based on multi-scale feature fusion is proposed.A variety of improved multi-scale feature fusion networks are used to improve the YOLOv3 network to improve the feature extraction capability and utilization efficiency of the network,thereby improving the detection accuracy,positioning accuracy and operating speed of YOLOv3.(2)Aiming at the problems of low detection accuracy and insufficient positioning accuracy of YOLOv3,a pedestrian vehicle detection algorithm based on the attention mechanism is proposed.Use a variety of improved attention networks to improve the YOLOv3 network,so that the network can filter out important features in a large number of features,avoid the interference of redundant and useless features,thereby improving the accuracy and positioning accuracy of YOLOv3 detection.(3)Aiming at the problem that current target detection algorithms cannot meet the accuracy,positioning accuracy and real-time requirements of pedestrian vehicle detection,a pedestrian vehicle detection algorithm based on multi-scale feature fusion and attention mechanism is proposed.Combining the improved multi-scale feature fusion network and the improved attention network,YOLOv3 is comprehensively improved to form the final algorithm of this article.Experiments on the BDD100 K data set show that the proposed pedestrian detection algorithm based on multi-scale feature fusion and attention mechanism increases the detection accuracy of medium targets and small targets by 4.67% and 7.65% respectively,compared with YOLOv3.In addition,the positioning accuracy is increased by 9.2%,and the running speed is11 FPS faster.The proposed pedestrian detection algorithm can realize the detection of pedestrians and vehicles in real road scenes.This paper contains 35 figures,8 tables and 63 references.