由于加入了深度相机点云数据，submap的信息量变得可观，单个subm

Due to the addition of depth camera point cloud data, the amount of submap information becomes considerable, and the amount of data of a single submap also becomes huge. During the loop detection, the newly observed data is directly matched with the key frame submap, and solving it is very time-consuming and laborious. The above loop detection based on the branch and bound algorithm is solved on the 2D submap. If you directly discard the data of the 3D point cloud and use the submap formed by the laser points alone, the accuracy of the loopback cannot be guaranteed, and it is also a waste of data information. Therefore, the 3D point cloud is stored in the submap in the form of descriptors. The descriptor not only greatly compresses the amount of original data, but also retains sufficient environmental information.

With the addition of deep camera point cloud data, the amount of information in submap becomes considerable, and the amount of data in a single submap becomes large, and it is time-consuming to match the newly observed data directly with the key frame submap during loopback detection. The loop detection based on the branch ingelimiting algorithm is solved on the submap of 2D. Directly discard the data of the 3D point cloud, use the submap formed by the laser point alone, the correct rate of the loop can not be guaranteed, but also a waste of data information. Therefore, for 3D point clouds are stored in submap in the form of a descriptor. The description sub not only greatly compresses the amount of raw data, but also retains sufficient environmental information.

Due to the addition of depth camera point cloud data, the amount of information of submap becomes considerable, and the amount of data of a single submap becomes huge. In loop detection, it is time-consuming and laborious to directly match the newly observed data with the key frame submap. The loop detection based on branch and bound algorithm is solved on 2D submap. If the data of 3D point cloud is directly discarded and the submap formed by laser points is used alone, the accuracy of loopback cannot be guaranteed, and it is also a waste of data information. Therefore, 3D point clouds are stored in submap in the form of descriptors. The descriptors not only greatly reduce the amount of original data, but also retain enough environmental information.<br>