Research on Loop Closure Detection Based on Semantic Information

[paper]

Abstract

In recent years, Artificial Intelligence has important research value in various fields. Among them, robotics, which is a representative of high technology, has also become a key research area. The robot must first be able to autonomously sense the surrounding environment information and accurately locate its own position, which is also the prerequisite for the robot to be able to autonomously navigate and move. Therefore, Simultaneous Localization And Mapping (SLAM) is an important research topic. However, to perform high precision SLAM, a high-performance loop closure detection module is essential. The loop closure detection module can significantly reduce the localization drift caused by the front-end odometry. But it is very challenging to design a loop closure detection algorithm with high detection accuracy under high recall. The method proposed in this thesis uses lidar as the main sensor to quickly and robustly detect whether there is a loop closure by detecting semantic information in the environment and then extracting global descriptors and local descriptors.

The work of this thesis mainly includes the following aspects: (1) This thesis proposes a loop closure detection method based on object level using only LiDAR data in an urban driving environment. This method innovatively extracts the point cloud descriptor from the semantic information in the scene, and enhances the point cloud descriptor’s ability to describe the object. At the same time, the method in this thesis improves the efficiency of detecting loops based on the idea of searching from coarse to fine. (2) Combining graph theory and semantic information, this thesis proposes two graph-based descriptors, one belongs to the global descriptor and has the necessary rotation invariance for loop detection, which is used to construct kd-tree and quickly find and loop candidate point clouds similar with the query one; the other belongs to the local descriptor, which is used to calculate the one-to-one correspondence relationship between the semantic objects between the two point clouds and handle the data association problem. (3) This thesis studies four other traditional loop closure detection methods and conducts a large number of experimental comparisons with the method proposed in this thesis on the large public KITTI dataset, and verifies many aspects of this method compared with other traditional loop closure detection methods.