Kinect-based 3D Indoor Environment Map Building For Mobile Robot

3D(three-dimensional)map is widely used in our daily life;for instances,autonomous navigation for mobile robot,medical imaging,virtual reality,and etc.However,the 3D map reconstruction for indoor use at present consumes too much time,and the information is not complete.Then it is difficult to be applied to actual robot navigation.In this manuscript,by extracting and matching of feature points,different positing solving methods are used under different environmental conditions,and then closed-loop detection and graph optimization are performed to the solved poses.Therefore,the 3D map with high precision is obtained and the point cloud data can be organized and indexed to perform the mobile robot navigation.This manuscript studies the real-time 3D reconstruction of indoor scene in computer vision based on the SLAM(Simultaneous Localization and Mapping)algorithm of mobile robot and the depth information obtained by a Kinect camera.It mainly includes the following works:(1)The Kinect camera and the Voyager II mobile robot are chose to construct our experiment platform.The Kinect camera is used to get 3D visual information,and the Voyager II mobile robot is used as the experimental hardware equipment.Then the software about 3D mapping and localization of mobile robot is developed within Microsoft Visual Studio 2010 and the open source software PCL(Point Cloud Library).(2)This manuscript proposes a comprehensive scheme for 3D color mapping and localization.First,the description of robot's pose and coordinate transformation were introduced.The feature points were detected and matched by using SURF(Speeded Up Robust Features)algorithm.And then corresponding algorithms were applied to calculate the visual odometer under different environments.On the basis of the robot's initial pose,the 3D map and update mechanism were established,and the initial location information was presented.By comparing the experimental results to theoretical values,the visual odometer and the scheme of mapping and localization were proved to be reliable.(3)For the problems of the accumulated error of visual odometer and the map's deformation and highly calculation cost when the map was large and the Kinect camera had worked for a long time,this manuscript presented a solution for the closed-loop detection,mapping and optimization based on key frames.By combing the similarity of look and distance in the closed loop detection method,this manuscript proposed a new method that consider the similarity and Euclidean distance together.Then,the closed loop detection under complex trajectory was analyzed experimentally to ensure the stability and consistency.(4)Based on the closed loop detection of key frames,the L-M(LevenbergMarquardt)algorithm was applied to optimize the map.Then,various indexes of the effects of key frames' mapping and trajectory before and after the optimization were analyzed experimentally.The results showed that the optimization improves the accuracy and performance of 3D mapping and localization.(5)In order to achieve the obstacle avoidance navigation and trajectory planning for mobile robot,the organization and spatial index method for the optimized 3D map point cloud data was improved.The scheme was proposed to improve the octree encoding,therefore to reduce the consumption of memory and to improve the retrieval efficiency.