Mobile Robot 3-D Scene Reconstruction In Semi-StructuredEnvironments

Mobile robot self-localization and 3-D scene reconstruction without using GPS information in unknown environment is a hotspot in the field of mobile robot problems. Mobile robot self-localization is the foundation for obstacle avoidance, navigation, and 3-D reconstruction has broad prospects in service robots, space exploration and disaster relief. This paper utilizes visual odometer to locate mobile robot in unknown environment, and achieves multi-view 3-D scene reconstruction with our Self-developed 3-D laser.This paper constructs a visual odometer system based on stereo camera. The system is composed of image feature point extraction, feature point matching, outlier rejection and pose estimation. An iterative image bucketing method is presented to maintain the number of feature matches at a low level. An inlier selection approach developed from the max-clique is used to improve the performance of RANSAC in the outlier rejection module.Due to the range of the visual sensor is limited, this research utilizes a self-developed 3-D laser based on Hokuyo 2-D range finder to perceive the surroundings. The calibration of each sensor module is done on the SmartROB platform,3-D point cloud is shifted with the pose provide by odometers. By the method of collecting 3-D laser data from different viewpoints, the complete information of the scene is achieved. In order to improve the efficiency of 3-D reconstruction, we utilize optical encoder for 3-D scene reconstruction, and use point cloud registration method for dynamic error compensation in scene reconstruction problems.The experimental Platform was built with ROS (Robot Operating System) on our SmartROB mobile robot. The indoor rooms, corridors and halls experimental results have represented and verified the effectiveness and practicability of our approach.