Recently, there has been a great interest in using autonomous land vehicles to build 3D maps of environment of uncertain information, which is increasingly concerned by the researchers. In fact, 3D map building, for autonomous robots, is a process, in which they model the surrounding grounded on information from perception. This article presents a technique for 3D map building based on data from localization sensors including inertial sensors and GPS and depth maps provided from a stereo vision system, within the context of autonomous robots. A list of experiments on 3D map building showed that the method brought forward in this article is efficient and practical.The prophase algorithms of 3D map building were presented at first involved camera calibration, image rectification and stereo matching, depth data gaining and so on. Based upon the dense disparity map from two 2D images of scene, the height of terrain could be accounted, and the quantity of data was reduced greatly by subdividing the whole scene into small subunits.In Chapter 3, the 3D scene was portrayed with OpenGL based on information of the subunits of a single frame. The results of visualization were improved by triangulation, pseudo-color methods and vector taken into account. These algorithms are indispensable preparation to the 3D map building.Chapter 4 presented a whole solution to 3D map building. Firstly, we discussed how to get a suit of world-based coordinates for small subunits. Secondly, data optimization was introduced and we defined block subunits and their initial values. Thirdly, we ascertained the regions of data in effect, and the height of terrain could be calculated based on block subunits. In the end, the results of these experiments were modified. As a result, the 3D map can meet the request of exactitude and reality. The terrain map can be applied to autonomous robotic systems or to automated car driving systems, for modeling the road, identifying obstacles and roadside features. |