Automatic Driving Vehicle Path Planning Based on Deep Convolution Neural Network

Autonomous vehicles need to keep a certain safe distance from obstacles. Aiming at the problem of over fitting phenomenon in obstacle recognition, which leads to the reduction of path search performance, a path planning method of autonomous vehicle based on deep convolution neural network is proposed. The two-dimensional environment model is used to describe the surrounding environment of the vehicle, establish the automatic driving structured road, and generate the virtual local target points. The deep convolution neural network is used to identify obstacles. The convolution layer maps the input data to the feature space, and the full connection layer maps the feature space to the sample label space through linear transformation, that is, the final output sample category, to complete the detection of virtual target points. Use the information of the target point to search the path towards the target, remove the obstacle node through the collision check function, judge whether the expansion node has reached the target area, and realize the path planning. The simulation results show that the average search time, the average number of search nodes and the number of successful searches of the automatic driving vehicle path planning method based on deep convolution neural network proposed in this paper are better than those based on genetic algorithm and SVM. The design method has significantly improved the search speed and efficiency.