Research On Key Technologies Of Automatic Driving Environment Perception Based On Cloud Computing

With the development of 5G communication technology and intelligent connected vehicle technology,the cloud-based automatic driving environment perception system can not only improve the driving environment perception range,but also save the vehicle-side computing power.This paper focuses on the key technologies of cloud-based automatic driving environment perception,which has important theoretical significance and engineering application value.The main research work of the thesis is as follows:The characteristics and limitations of the traditional vehicle environment perception system are analyzed.The overall framework of the automatic driving environment perception system based on cloud computing is studied.The cloud-end collaboration features and advantages of the framework are analyzed,and the key implementation technologies of each platform under the framework is further analyzed.A unified framework of license plate detection and license plate recognition is designed.A license plate detection and recognition fusion algorithm based on SSD(Single Shot MultiBox Detector)neural network model is proposed.The model is trained and tested by the open source license plate dataset CCPD.The accuracy,robustness and running performance of the algorithm are analyzed and verified.Based on the algorithm,the driving environment perception static scene is constructed in the cloud,which expands the scope of automatic driving perception.A safe driving route awareness algorithm based on convolution Social Pooling is proposed.The algorithm senses the safe driving route of the vehicle based on the historical trajectory of the surrounding vehicles.Based on this algorithm,an improved algorithm for integrating the driving intention information of surrounding vehicles is proposed.Through the training and testing of the algorithm on the open source NGSIM dataset,the influence of surrounding vehicles on the safe driving route perception is studied,the optimal structure of the convolution Social Pooling module in this algorithm is studied,and the driving intention information of surrounding vehicles is analyzed.The result proves that using the surrounding vehicle driving intention information can improve the accuracy of the sensing algorithm and improve the sensing effect for long periods.The common framework of distributed deep learning is analyzed.The data partitioning and model partitioning methods,communication mechanism and model aggregation method in distributed deep learning are studied.The algorithm flow and improvement scheme of model aggregation method are studied.In the framework of TensorFlow,a distributed deep learning algorithm is implemented,which effectively improves the convergence speed of the model.At the end of the thesis,a simulation scene of automatic driving is constructed,which realizes the safe driving environment perception of automatic driving based on the algorithm in the cloud,and achieves good results.