Obstacle Detection And Road Segmentation For Robots Based On Information Fusion

Mobile robots have been widely applied in the fields of military, civil application and science research, drawing a lot of attention from researchers. Navigation system is a popular and difficult subject in robotics, serving as a basic module for robot automatic system, it has great influence on mobility and intellection of mobile robot. Since single sensor can not fully understand surrounding environment, mobile robots usually employ several sensors and apply information fusion technology in order to enhance robustness of navigation system. This paper focuses on two key technologies of autonomous navigation system: obstacle detection and road segmentation, where information fusion technologies are applied.This paper firstly introduces the background, developments and the state of the art of mobile robots' research. Then robots system architecture and key application technologies, including general theories of multi-sensors information fusion, obstacle detection and road segmentation are presented.We use two kinds of sensors, laser range finder (LADAR) and camera to carry out joint calibration. The whole procedure goes as follows: a calibration board is put in front of ladar and camera, acquiring the scene data synchronously by these two sensors with different board position.. Feature points on the calibration board are picked up to perform calibration.A novel method fusing information from ladar and camera for detecting obstacles in grass is proposed. Use Expectation Maximization algorithm to learn the Gaussian Mixture Model which represents the distribution of shape features. Then 3-D points of candidate obstacles in the scene can be obtained. Meanwhile, mean-shift algorithm is applied to the corresponding color image to obtain the region information of the scene. With the result of joint calibration of the 3D-ladar and the camera, those candidate obstacle points are projected to the segmented image and fusion is then performed to make the final decision.At last, this paper proposed a road segmentation algorithm. We project 3D-ladar points onto corresponding pixels in mean-shift segmented image according to the result of joint calibration, then use height to distinguish road region from background.