Research On Visual Loop Closure Detection Based On Spatio-Temporal

With the expansion of exploring,the robustness of simultaneous localization and mapping in large-scale environment is particularly important,and reliable loop closure is the most important and critical issue of building robust SLAM.Loop closure detection is that the robot judging whether its current position is in the region it has visited,and then to update the map or not,which allows the robot to reduce the uncertainty associated with the state variables that define the robot pose and the map.However,the shortcomings and inevitable calculating errors of modeling in visual information acquiring,describing and matching cause great difficulty in the mobile robot to extract the precise loop closures.The visual loop closure detection is still one of the most challenging problems in outdoor environments.Our goal is to deal with the data redundancy and computational complexity of loop closure detection by researching key frame extraction and detection of loop closure information.The main works are as follows:Firstly,the Kinect is calibrated and registered to correct the image distortion and align the color and depth map.Secondly,in order to extract the key frames from the visual scene effectively,spatio-temporal slices is proposed based on the analysis of key frame extraction technique,and its slope is used to represent status of motion.On this basis,the nearest neighbor pixel-matching method is used to determine the key frame which can accurately describe the motion state of the robot.The experimental results have shown that the proposed method algorithm has good performance both in the calculation accuracy and processing speed.Then,to deal with the computational complexity and perceived ambiguity,the shortest path spatio-temporal slices algorithm is proposed.By using the Dijkstra's shortest path algorithm to remove the remote frame from the current frame,we can solve the problem of visual confusion caused by the spatio-temporal slices,and then select the correct loop closure information by using spatio-temporal slices.In order to increase the closed loop constraint,Dijkstra's shortest path is used to select loop closure information once again.Finally,this paper presents an algorithm of shortest path spatio-temporal slices based on contour extraction,which can solve the problem of selecting loop closure information in the single scene.Aiming at the problem that the scene information of color image is affected by the illumination and color information,the depth information of Kinect is processed.The depth image is filtered to fill the hole.The connected region labeling method is used to extract the contour of binary depth image in order to find the most obvious region in the scene,and spatio-temporal slices is used to extract the slice pixels of the region,so as to avoid extracting the region with too much uniformity.The superiority of the algorithm is shown by comparing with RGBDSLAM.The contributions of this work not only improve the efficiency and accuracy of loop closure detection,but also provide some reference for improving the accuracy and reliability of the robot navigation.