Research On Environment Perception And Control Method Of Autonomous Vehicle Based On Deep Learning

Environmental perception and control is the key technology to achieve the autonomous vehicle.In recent years,deep learning technology has developed rapidly.It has high accuracy and robustness in machine vision,natural language processing and control operation.In this paper,the problem of multi-scale object detection and end-to-end control in unmanned environment perception are studied based on deep learning technology.Firstly,in order to ensure the accuracy of the deep learning algorithm and the applicability of the model in the field of autonomous vehicle.This paper establishes a special data set for in-depth learning of autonomous vehicle.After effective screening and labeling,the data set in this paper contains about 400,000 targets.It basically covers all kinds of driving conditions,and has the characteristics of comprehensive scenes and high accuracy.Secondly,in the aspect of autonomous vehicle environment perception,the current classical target detection framework and the mainstream multi-scale target detection algorithm are studied and analyzed in this paper.Based on the analysis of the results,the relationship between the spatial resolution and the semantic level of convolution features is comprehensively processed.A multi-branch deep learning network with different spatial resolution and the same semantic level features is proposed from the perspective of the structure of deep learning network.The network proposed in this paper is compared on several representative public datasets.The validity of the proposed algorithm is verified.Thirdly,based on the existing research on end-to-end control of autonomous vehicle,this paper comprehensively analyses the characteristics of end-to-end control models in different periods,and combines the multi-branch network model structure with good feature extraction ability,reformed the multi-branch network model,designed an end-to-end control model based on multi-branch network model and implemented it in the control of driverless vehicles.Finally,in order to prove the applicability of the proposed method in the field of autonomous vehicle,this paper carries out relevant experiments on the self-built data set of autonomous vehicle.The experimental results show that the proposed method can better solve the problem of environmental awareness and control of autonomous vehicle.