A Semantic-Based Loop Closure Detection of 3D Point Cloud

Compared with loop closure detection based on vision, loop closure detection based on 3D point cloud is more robust against the changes in the external environment, and therefore has attracted more and more research interests. However, due to the sparse and discontinuous characteristics, the point cloud is susceptible to noise points and the occlusion, which renders the loop closure detection task challenging. Here, we proposed a semantic-based loop closure detection method, which explores semantic objects and their topological for loop closure detect. The semantic object obtained through the semantic segmentation model improves the descriptors representation. In this work, the main axis direction is determined through the semantic PCA (principal component analysis) algorithm, and the local column shift is applied to reduce the influence of noise points and the occlusion. Finally, the similarity calculation is performed on the semantic images. The feasibility of proposed method is evaluated through the KITTI dataset and the results show that the prosed method outperforms the state-of-the-art method. Our code is available at: https://github.com/fanvanfu/PCA-SSC.git.