Built Three-dimensional Terrain Robot-based Laser Radar Obstacle Detection And Grass

With the development of science and technology, the modern robots have been widely used from manufacturing industry to many different fields such as military application, civilian application, service industry and scientific research. A robot is capable of performing some disgusting, risky or monotonous tasks assigned to human beings. In the area of robots, laser radar is a very important sensor and nearly every robot uses it. This paper used a three-dimension (3D) data collection system that was rebuilt from a 2D laser radar to research two important robot moving issues that were the outdoor environment rebuilding and the detection of obstacle in grass.The first part of the dissertation was to resolve how to build a 3D data collection system using a 2D laser radar. Compared with 2D laser radar, the 3D laser radar is more visual and has larger scale of view, but it is usually very expensive. The dissertation used the stepper motor and MSC-51 control part to drive the 2D laser radar rotate by the horizontal axis to collect the 3D data. And the precise 3D information was gained by calibrating.The second part of the dissertation was to resolve the outdoor environment building using the 3D data collection system. Because the range data collected by laser radar was obtained relative to the radar reference frame, we must registered scans from different positions into the world coordinate by using the laser radar position and attitude. So the ICP registration algorithm was proposed, and it was improved in reducing the operation time, deleting the non-overlapped area to improve the precision of the algorithm and eliminating the local minimum problem. A backward registration method was proposed to reduce the accumulation error of registration of multiple depth pictures. At last, a 3D map was built using the OpenGL.The third part of the dissertation was to resolve the detection of obstacle in high grass. When the robot moves in a field of grass, the grass can't be seen as obstacle. So the grass must be removed to detect the true obstacle. The flat obstacle was extractedin the environment by using the "max-value filter" and picking the small absolute value of the second derivative of the range value. And the small wrong obstacles were eliminated by the data of multiple scan planes and multiple scan pictures. At last, the kind of cylindrical obstacle was detected using the improved Hough transformation method.