Vehicle steering control method based on machine learning

Abstract

The environmental perception and control are the core issues of driverless vehicles. Considering the weak signal of global positioning system (GPS) and the difficulty of road information acceptance in rural area, the DIPLECS autonomous driving datasets is selected. The roads in dataset are classified by support vector machine algorithm (SVM), and established into the road template library. Then, on different roads, images and steering direction are used as the input and labels of the training set for back propagation neural network (BPNN) model. Finally, the effect of the model is verified on the video clip. The results show that the accuracy of road classification reaches 99.8% through SVM, the shortest time consumption of a single frame is 0.9 ms. The highest accuracy of steering prediction is 94.5%, and it only takes 0.224 ms to detect a frame. The algorithm meets the requirements of accuracy and real-time, and it can effectively assist and warn the drivers or automatic driving systems.