Multithreaded Edge-Assisted Visual SLAM with Keyframe Backup Mechanism

Simultaneous Localization and Mapping (SLAM) has rapidly advanced on mobile devices, particularly camera-based Visual SLAM, which enables spatial perception and autonomous positioning by processing continuous image data. Due to its high memory and processing demands, it is challenging to deploy and execute continuously for a long time on mobile devices. The edge-assisted architecture that mitigates resource constraints by offloading heavy tasks to an edge server becomes optimal for settling the problem. However, existing studies suffer from high data synchronization delay, which is handled in the tracking module, resulting in prolonged tracking interruption and poor system robustness and accuracy. Based on a typical edge-assisted Visual SLAM system, we analyze the impact of the data synchronization process and propose a new multithreaded tracking solution with a keyframe backup mechanism. Verifying through two standard datasets, we evaluate our system’s robustness and localization accuracy. The results show that the proposed system reduces tracking interruption by up to 88.1% and significantly improves the coverage, a critical robustness metric of the SLAM system, by up to 25.5%. Additionally, our proposed solution significantly improves localization accuracy, especially in rotation scenarios, by up to 26.7%.