Research On Vehicle Pedestrian Detection Method Based On Deep Learning

As one of the hot research contents in the direction of computer vision in recent years,vehicle and pedestrian detection technology is particularly widely used in the fields of intelligent transportation,assisted driving and automatic driving.It can avoid the occurrence of traffic accidents as far as possible and better ensure the safety of drivers,which has great research significance in real life.However,due to the different scales of vehicle and pedestrian targets in the real traffic scene and the occlusion of each other,as well as the influence of complex traffic road environment,various current target detection methods have problems such as low detection accuracy,slow speed and missed detection of small targets and occlusion targets.Therefore,vehicle and pedestrian detection is still a very challenging problem.The rapid development of deep learning has brought new development opportunities for vehicle and pedestrian detection.In view of the existing problems in vehicle and pedestrian detection,this paper chooses the SSD model with good performance in detection accuracy as the basis,and takes advantage of the fast speed of the network model to propose an improved vehicle and pedestrian detection method of SSD.Based on SSD,this paper proposes a method of vehicle and pedestrian detection based on dual attention feature fusion.The dual attention modules of self-coding spatial attention and channel attention are applied in the feature fusion strategy of bi-directional feature pyramid to improve the detection performance of small-scale targets and partially cross-occlusion targets.Dual attention module is used to highlight the key effective features of the target and suppress the expression of irrelevant features.This feature fusion method can enhance the expression of features.Replace the regression loss of the frame,and predict the loss of the boundary box coordinates through the CIOU loss,improve the accuracy of the positioning of the detection box,and achieve better regression of the frame.The focus loss function of Focal Loss can alleviate the problem of data imbalance and make the classification more accurate.In addition,a vehicle and pedestrian detection network structure based on Res Next50 SSD is proposed.Res Next50 network is used to replace the backbone network VGG of the original SSD algorithm,and the loss function is replaced by the previously proposed feature fusion strategy to effectively improve the detection performance of vehicle and pedestrian targets.In this paper,the improved method is experimented on the KITTI dataset and compared with other methods.The experimental results show that the improved method can effectively improve the detection accuracy of vehicles and pedestrians in traffic scenes,and the improved mean accuracy MAP reaches 79.6%,which is 4.1 percentage points higher than the original SSD.It reduces the small targets of vehicles and pedestrians,and the missed detection of mutual occlusion targets,and the detection effect is good.