3D Map Building Based on Stereo Vision

Abstract

In this paper, we present a 3D map building method based on stereo vision for autonomous land vehicle(ALV). We achieve the 3D coordinates under the camera coordinate system(C-CS) from the disparity map firstly. Then we develop the conversion between the C-CS and the ALV coordinate system(ALV-CS) to acquire the local 3D map which reflects the actual scene basically. And the method is highly real-time and meets the need of ALV navigation. Furthermore, combined with the binocular stereo vision system, the error of the method is analysed based on the 2D prediction error from camera calibration and the disparity error from stereo matching and the error item that describes the depth error between the built data of the scene point and the actual one is computed. The experiment to compensate the built 3D map is done and the result shows that the item can improve the precision of 3D map effectively. Then combined with information from INS and GPS, a global 3D map that reflects the actual scene, is built based on local 3D map.