Steering Wheel Rotation Angle Prediction Based on VGG-16 and Data Augmentation

Abstract

The deep learning technology based on convolutional neural network in the field of automatic driving makes automatic driving have extremely high practical application value and scientific research value. With the continuous improvement of network real-time and edge computing, autonomous driving technology will continue to show its powerful strength to liberate people's hands in the process of driving. Therefore, controlling the direction and angle of the steering wheel rotation according to the picture captured by the camera in front of the car is the core problem needs to be solved by automatic driving. We design and implement a steering wheel rotation angle prediction system for autonomous driving. By collecting road information on the simulator, data balance analysis is carried out on the collected road picture information. We designed related algorithms such as horizontal flipping of data balance processing, angle processing of bilaterally collected pictures, and random elimination of 0-angle information. For scenarios such as the correlation conditions in the prediction process of the actual vehicle rotation angle, we have done a complete and scientific data augmentation experiment, and trained an excellent prediction network model. From the results of the road test, it can make excellent performance in multi-scenario and complex road conditions, and can more accurately predict the trajectory and direction of the vehicle to be driven. It laies a certain theoretical foundation and accumulated practical experience for the development of autonomous vehicles under smart transportation.