Reducing Communication Consumption in Collaborative Visual SLAM with Map Point Selection and Efficient Data Compression

Efficient data communication is a challenging problem for Collaborative Visual Simultaneous Localization and Mapping (CVSLAM), particularly in bandwidth-limited applications. To resolve this problem, we propose a communication load reduction method. We first propose a map point culling strategy by considering maximum pose-visibility and spatial diversity, to eliminate redundant map information in CVSLAM. Then, we employ a Zstandard (Zstd) compression algorithm to compress visual information so as to reduce the required communication bandwidth. To exhibit the efficiency of the suggested approach, we implement this method in a centralized collaborative monocular SLAM (CCM-SLAM) system. Extensive experimental evaluations indicate that our method can reduce communication overhead by approximately 49% while maintaining map accuracy and real-time performance.