Forward Vehicle Detection Based on Incremental Learning and Fast R-CNN

Recently the research of vehicle detection is mainly through machine learning, but it still has low detection accuracy problem. With the study of researchers, using deep learning methods of vehicle detection becomes hot. In this paper, a selective search method and a target detection model based on Fast R-CNN are used to detect vehicle. The strategy optimizes the model by preprocessing the sample image and the new network structure. Firstly, the experiment uses the public KITTI data set and self-collected BUU-T2Y data set, respectively, for training validation and test. Secondly, based on the original data set, the experiments go on through incremental learning, combining the KITTI dataset with the BUU-T2Y dataset. The experimental results show that the proposed method is superior to the result of multi-feature and classifier detection in terms of accuracy. To a large extent, the proposed method solved the problem of missing vehicle for detection and improved the accuracy of vehicle testing and robustness.