Mobile Robot 3D Scene Reconstruction Based On Laser Scanning

Effective recognition and perception in 3D environment are two essential tasks for mobile robot's navigation and exploration. Compared with structured environment, unstructured environment is more uncertain, unstable and unpredictable. The objective of this thesis is to quickly, efficiently and independently accomplish 3D unstructured environment's scene reconstruction based on the autonomous mobile robot system which has the ability of perception and recognition.An automatic system of reconstructing 3D unstructured environment is presented in this thesis. It consists of a laser ranging system and an autonomous mobile robot system. The software can provide a series of functions such as real-tine registration and reconstruction, which demonstrate the algorithms' validity and practicability.This thesis presents a novel and effective ICP (iterative closest points) algorithm based on endpoints of lines as a solution to scene registration in different viewpoints. The line feature can be extracted quickly and accurately by using the point clouds' own order. The endpoints of line features can demonstrate the key information of scenes to a certain extent. This thesis presents ICP algorithm based on endpoints of lines to align different 3Dscans. Further restrictions for the matching points are brought forward to improve the registration accuracy. The presented method is tested to be an efficient and independent solution to the problem of 3D scene matching in indoor laboratory environment.The order of multiple scan matching can affect the effectiveness and accuracy of scene reconstruction. By analyzing the relationship among different viewpoints, this thesis presents a novel reconstruction method based on the core scene. For the redundant data in the overlapping area between neighbored scenes, a reduction strategy based on a 3D-grid-division method is presented. Different strategies of reconstruction are adopted in the experiment of aisle environment's reconstruction. Experiments results and data analysis demonstrate our method's validity and practicability.