Research Of Simultaneous Localization And Mapping Based On Kinect

With the increasing demand for mobile robots and the increasing complexity of its working environment, more and more attention has been paid to the needs of locating and navigating of mobile robot in a completely unknown environment. Simultaneous localization and mapping is the basis and key technology of realizing highly intelligent mobile robot, which is able to map and locate mobile robot in an unknown environment. As a kind of equipment which can effectively acquire the information of environment texture and structure, vision sensor has the feature of light weight, small volume and high cost performance, and has been widely used in computer vision and mobile robot.In order to solve the problems of simultaneous localization and mapping in an unknown environment, research on the image registration, loop detection and graph optimization based on Kinect sensor was conducted in this study, which realized the accurate estimation of Kinect motion in an unknown interior environment and the map construction of the environment. The main research contents of this study are as follows:(1) Based on the data characteristics of depth image and the analysis of noise error of original image collected by Kinect, bilateral filter was used to filter the depth data to improve the mapping accuracy.(2) The inter frame registration based on sparse image features and the relative motion between frames was studied. On the basis of the commonly used methods of image feature extraction and description, different methods of feature matching was studied and the influence of the matching results on the motion estimation accuracy under different threshold conditions was analyzed. In the meanwhile, in order to achieve higher estimation precision of inter frame motion, a bidirectional PnP motion estimation method based on RANSAC were adapted in this study. Experimental results show that the combination of these two methods can improve the estimation accuracy of frame motion effectively.(3) According to the cumulative error of motion between adjacent frames, loop detection technology was studied. Based on the principle of loop detection, the loop closure detection based on the information of image registration and appearance was realized. The experiment of the influence of loop detection on the final Kinect trajectory was carried out, which show that the loop detection can effectively reduce the cumulative error of Kinect trajectory.(4) Aiming at the problem of describing the constraint relation between Kinect pose and position, a optimization method based on pose graph was studied. The theoretical basis of graph optimization, including the construction of pose diagram, the theory of graph optimization, was introduced. A method of using the reprojection mean error estimated by relative motion as the basis for the construction of the side information matrix was proposed in this study, which can effectively reduce the influence of large margin constraints on the optimization results. Then the detailed steps of solving SLAM optimization problem by using the g2o general graph optimization framework were introduced.(5) On the basis of the aforementioned process algorithm, the performance of the algorithm was tested by using the public data set. The real-time performance and the accuracy of the algorithm were analyzed. Through the constant adjustment of the relevant parameters of the algorithm, the accuracy and speed of the algorithm had reached the expected requirements. At the same time, Kinect was used to carry out the indoor SLAM experiment, and the performance of the estimation accuracy was qualitatively measured by the effect of map generation.