Hardware-dependency-reduced visual slam

Abstract

To address the poor real-time performance and hardware dependency of traditional visual Simultaneous Localization and Mapping on embedded platforms, this paper proposes a cloud-native collaborative framework enabling low-latency data transmission and cloud-based pose estimation. Experimental results demonstrate stable frame rates without map fragmentation or tracking failures in scenario testing. The architecture enhances environmental adaptability and computational robustness for visual Simultaneous Localization and Mapping on compact unmanned platforms, providing a high precision solution for resource-constrained scenarios.