Research On Key Technology Of Unmanned Cruise Vehicle Autonomous Driving

The key technology of unmanned vehicle autonomous driving is one of the hot topics in academic research all over the world,and its potential commercial value leads generations of researchers to explore it wider and deeper.Once the automatic driving technology matures,the intelligent transportation and smart city will subvert the human understanding of traffic travel,and affect every aspect of human life.Therefore,the research on the key technology of unmanned vehicle in this paper has important practical value.Firstly,the paper studies the transformation relationship of each coordinate system in the field of image ranging.At the same time,the advantages of algorithms in image preprocessing,camera calibration,stereo matching and post-processing are analyzed experimentally.And then,the unmanned vehicle binocular ranging scheme based on the cross-window cost function aggregation stereo matching algorithm is given,and the feasibility and real-time performance of the algorithm in the cruising unmanned vehicle condition are verified.Next,the paper studies various image feature matching algorithms,gives the rotational adaptive image mosaic algorithm based on ORB,the algorithm satisfies the translation invariance,rotation invariance and certain scale invariance of image stitching in the process of automatic driving of unmanned vehicles.And it proves the superiority of the rotational adaptive image mosaic algorithm based on ORB in the field of unmanned driving by comparing the experiment.Finally,the paper studies the basic structure of the full convolution neural network,analyzes the functions of each part of the module,and gives an optimization model based on FCN-8S,which adds dilated convolution,unpooling and deconvolution on the structure of FCN-8S.The paper also gives an optimization model based on ResNet101,with blocks such as atrous spatial pyramid pooling and CRF.The model given in this paper is compared with the classical models such as FCN-8S,SegNet and DeepLab,and the superiority of the semantic segmentation algorithm given in this paper is verified by experimental comparison.