Loop Closure in 2D LIDAR and RGB-D SLAM

According to the advantages of different sensors, it can improve the speed and accuracy of Simultaneous Localization and Mapping (SLAM) based on multi-sensor information. However, with the accumulation of time, the error will gradually increase due to the factors of environment and hardware in Mapping. Therefore, the paper proposes a loop closure algorithm based on the combination of 2D LIDAR and RGB-D camera. The algorithm will create a grid map by the use of 2D LIDAR, and use a coarse-to-fine matching method on the candidate region. Firstly, it will get the RGB-D image of the current location and compare the global descriptors of the RGB image using the similarity evaluation based on distance. In the initial screening stage, part of the RGB image in the map storage with a high similarity to the RGB image at the current location will be selected. Secondly, use local descriptors to compare and extract the most similar RGB image, thus establish the corresponding relationship between the RGB images. Finally, the depth image corresponding to the RGB image is segmented through the local descriptors, and adjust the relative pose according to the point cloud matching technology to obtain a robust loop closure algorithm.