Research on Generalized Error Control Mechanism of Monocular Vision Ranging Method
Lixia Xue, Meian Li, Aixia Sun, Xiaoxin Jin, Yongan Zhang · Journal of Physics Conference Series · 2020
Abstract Distance perception is the basis and necessary prerequisite of environment perception, attitude perception and obstacle avoidance of intelligent vehicle and unmanned vehicle. Monocular vision ranging method is one of the mainstream distance sensing methods at present. In order to improve the accuracy of monocular vision ranging, a monocular vision ranging method based on machine learning is proposed. The monocular vision ranging method studied in this paper has the advantages of high accuracy, simple training data, simple ranging formula and can be explained, but the uncontrollable generalization error is one of the disadvantages of this method. Therefore, in order to explore the relationship between generalization error and training error, according to the monocular vision ranging method based on the pinhole imaging principle and a large number of measured data, this paper uses the polynomial method in the curve fitting toolbox of MATLAB to fit the functional relationship between coordinates and image distance, so as to obtain the model parameters. Finally, the threshold value of the optimal model is 0.5%, 49 training data and 21 test data. The extreme value of the ratio of the generalization error to the threshold value of the training error is 1.68, which can be used to control the generalization error of the monocular vision ranging method.