Visual Data Compression Approaches for Edge-Based ORB-VSLAM Systems

Visual Simultaneous Localization And Mapping (VSLAM) is a technique that enables robots to localize and estimate their pose by reconstructing a map structure of the surrounding environment. However, VSLAM encounters significant computational loads, resource consumption, and bandwidth constraints. Due to these issues, offloading the entire VSLAM architecture onto a mobile device could be impractical. To reduce weight, lower energy consumption, and keep the size of mobile devices small, VSLAM can be efficiently deployed by offloading computationally intensive tasks to cloud platforms. This requires efficient visual data compression methods to reduce the data load, bandwidth, and latency. This paper proposes two implementation approaches for integrating visual data compression and decompression with ORB-VSLAM systems. The proposed systems are compared to counterpart JPEG-based systems, and the results show significant improvement for both approaches.