Study On Vehicle Obstacle Avoidance System Based On Deep Learning

With the continuous development of social economy and the continuous improvement of living standards,car driving has become the primary choice for people to travel.However,long-term fatigue driving not only has hidden dangers to traffic safety,but also affects people's travel experience.If an unmanned driving system is installed on a motor vehicle,it can not only provide auxiliary control for the driver,it may even liberate the driver from the tedious and tired driving activities,so the use of deep learning theory to study automatic driving systems has been It has gradually become one of the global research hotspots.The key technologies of autonomous driving can be divided into three parts: perception,planning and control.The perception part is the most widely used technology field of computer vision.It mainly refers to target detection and target tracking.With the continuous improvement of deep learning theory,computer perception technology has been greatly developed.As a development trend in the future,intelligent vehicle is one of the first problems to be solved.Therefore,a set of vehicle obstacle avoidance system is designed based on deep learning theory.This system mainly includes three modules:detector,tracker and target distance calculation,and each algorithm module is deployed in the vehicle embedded system and server.In order to make full use of the deep learning model,the entire software system is designed as a client-server model;the embedded client with image acquisition function is responsible for signal acquisition,pre-processing and image data transmission,and the server will calculate based on the received data The final result,and return it.In order to ensure the detection effect of the detector network on the target vehicle,and solve the problem of uneven distribution of degraded samples,the embedded adversarial network system isused for offline training.Using real-time positioning and map reconstruction methods to estimate the distance between the target and the driving vehicle.By capturing the Sift features of the same vehicle in different image frames,the change of camera pose is determined.The image depth map is reconstructed according to the obtained relevant parameters,so as to estimate the target distance by using the relevant parameters of the camera.The calculation results are given according to the running results of the detector and tracker in the experimental phase and their respective time costs.The experimental results in this paper show that the method can provide detailed vehicle target distance information for the vehicle obstacle avoidance system,and the design system can provide a certain guiding significance for the subsequent research on automatic driving problems.