Research On Vehicle And Pedestrian Vision Object Detection Methods In Complex Scenes

In recent years,artificial intelligence technology has developed rapidly,and it has been widely used in all walks of life.Transportation industry is an important industry in China,and the important application of artificial intelligence technology in its industry is automatic driving.The self-driving car collects road information through the camera,identifies obstacles,such as vehicles and pedestrians,through the object detection algorithm,and finally makes subsequent motion decisions according to the obstacle information.However,due to the complexity of the actual road conditions,the existing algorithms have an impact on the recognition of vehicles and pedestrians,resulting in low recognition accuracy.For example,it is difficult to identify other pedestrians,vehicles,pedestrians and other targets in bad weather such as fog,rain,snow and so on.Therefore,in order to solve the problem of low accuracy of vehicle-pedestrian target detection in complex scenes,this paper proposes CSPU-Net image denoising algorithm and SGWYOLO target detection algorithm,and the specific work is as follows:(1)Aiming at the problem that bad weather in complex scenes will reduce the quality of video images captured by cameras,cause blurred targets in videos,and affect the efficiency of target detection algorithms,this paper proposes a CSPU-Net image denoising algorithm integrating channel fusion module and channel attention mechanism,and introduces MS-SSIM loss function.The denoising performance of the network can be significantly improved,and the color distortion after denoising can be better dealt with.The performance of the algorithm is verified by several data sets,and the denoising performance is greatly improved.(2)In order to solve the problems such as large occlusion and difficult identification of long distance small targets,a multi-scale real-time target detection algorithm SGWYOLO with Transformer is proposed.First,YOLOv5 s is prone to missing targets in dense vehicle and pedestrian detection scenarios.In the algorithm,the proposed SGWin Transformer V2 backbone network is used to replace the original backbone network.Secondly,CBAM module is introduced to make the model pay more attention to the region where the small target is located.Then,in the part of loss function,SIo U is used to optimize the loss function for the positioning of small targets,and the context information extraction module is added in front of each CSP module to increase the context information extraction capability of the model.By comparing the ablation data,we can see that the overall performance of the proposed algorithm is better than other target detection algorithms of the same category.(3)The combined improved image denoising algorithm and object detection algorithm are tested in real complex scenes,and the results of vehicle pedestrian detection using only the object detection algorithm and the two algorithms at the same time are compared,and the advantages of the combination of image denoising algorithm and object detection algorithm are verified.Based on the above,this article analyzes and researches the key technical issues of vehicle and pedestrian target detection under complex scenarios such as severe weather and road congestion,proposes reasonable solutions to obstacles brought by occlusion,density,and complex weather factors to intelligent detection for autonomous driving,and has certain reference value for future autonomous driving needs.