Binary Image Fingerprint: Stable Structure Identifier for 3D LiDAR Place Recognition

Place recognition is considered as an effective strategy to reduce robot drift errors. In this work, a place recognition method that uses binary features to match loop closure frames is proposed for 3D LiDAR. The method extracts important information from structural feature matrix by image compression, and the resulting binary matrix with small size is used as a unique identifier for each frame of the point cloud. This matrix is called the binary image fingerprint (BIF). The Hamming distance between two binary image fingerprints is used for similarity matching when performing place recognition. The use of logical operations to match loop closure frames shows high efficiency. The proposed method has been extensively experimented on KITTI, NCLT, and MulRan datasets and widely compared with state-of-the-art methods. The experimental results show that the proposed method exhibits superior performance under most sequences. Notably, the proposed method has high robustness to sparse point clouds.