Research On Loop-closing For Robot SLAM Based On RGB-D Images

With the advanced development of mobile robot technology, SLAM for a robot incomplex environments has become the key research area in the field of mobile robottechnology. How to estimate the robot’s current position is visited in an fast and accurateway is the key problem of the mobile robot in sore need of to be solved. Therefore, inorder to make the robot accurately determine the current area has visited, and overcomethe result affected by the environment of existing methods, with analysis of the existingapproaches, take advantage of the idea with combining the depth information with loopclosing detection, the paper proposes an mobile robot indoor loop closing method whichbased on RGB-D information. And focused on the following aspects.Firstly, put forward front region extract algorithm based on RGB-D information.Since the depth information can directly reflected the distance information of each objectin the scene, according to the different distance information, can simultaneously extractdepth and color information of the main object in the scene, implement extract the frontregion.Secondly, research contour extraction and optimization method, and match thecontour. As the data set contains a large amount of scenes information, in order to achievereal-time match, need to adopt fast matching method. Also the depth front regioninformation includes only the main shape of the object, so adopt the method by extract thecontour and matching. Using the advantage of RGB-D information, mixed the contourextracted by color and depth information, realizes contour optimization, last researchcontour matching algorithm research, achieves contour match.Thirdly, proposed a closed-loop detection method based on RGB-D information.Based on contour match selected the similar scenes, research for the similarity of scenes.Firstly, study image feature extraction algorithm, and improve the PCA-SIFT algorithm.Secondly, the study of closed-loop algorithm based on visual detection, form a visualdictionary based on extracted feature descriptor, and calculate the similarity betweenscenes, identify sequences of the same scenes, in order to achieve closed-loop detection. And improve the formation of visual dictionary.Finally, experimental results show that the proposed method is feasible and effective,and analyze the experimental results.