Real-time Self-calibration Method of Vehicle Camera Attitude Based on ANN

Abstract

Abstract The research of the monocular vision ranging method has great research value in the field of intelligent vehicles. Camera calibration is an important step in a monocular vision ranging model. The accuracy of camera calibration directly determines the accuracy of ranging. Because the camera attitude may change during vehicle driving, it will have a certain impact on the accuracy of the ranging algorithm. Therefore, based on a monocular ranging model and focal length calibration method based on geometric transformation, a real-time camera attitude self-calibration method based on the width measurement model and ANN model is proposed in this paper. Firstly, the method takes the wheel width of the known vehicle type as a fixed value, takes the contact points between the two wheels of the front vehicle and the ground as the feature points, and brings them into the width measurement model to form constraints. However, the single formula formed by a single pair of feature points cannot solve the camera inclination of multiple dimensions at the same time, and it is difficult to solve the analytical solution of the complex formula. In this paper, the ANN model is proposed to fit and solve the formula, and a large number of data sets are obtained through experimental point distribution to train and test the model. The results show that the ranging accuracy of using this method to calibrate the camera tilt angle is unchanged compared to the previous measurement accuracy using sensors, meeting the ranging requirements