A Multi-Channel Descriptor for LiDAR-Based Loop Closure Detection and Its Application

Simultaneous localization and mapping (SLAM) algorithm is a prerequisite for unmanned ground vehicle (UGV) localization, path planning, navigation, which includes two essential components: frontend odometry backend optimization. Frontend tends to amplify the cumulative error continuously, leading ghosting drifting on results. However, loop closure detection (LCD) can be used address this technical issue by significantly eliminating error. The existing LCD methods decide whether exists constructing local or global descriptors calculating similarity between descriptors, attaches great importance design of discriminative effective measurement mechanisms. In paper, we first propose novel multi-channel (CMCD) alleviate lack point cloud single information in power scene description. distance, height, intensity encoded into three independent channels shadow-casting region (bin) then compressed it two-dimensional descriptor. Next, an ORB-based dynamic threshold feature extraction (DTORB) designed using objective 2D describe distributions clouds. Then, DTORB-based method rotation-invariance visualization characteristic descriptor features overcome subjective tendency constant ORB extraction. Finally, verification performed over KITTI sequences campus datasets Jilin University collected us. experimental results demonstrate superior performance our state-of-the-art approaches.