Large-Scale Crowdsourced Mapping with EdgeSLAM

Indoor navigation complexities stem from the extensive setup and reliance on existing infrastructures. Recent advancements have explored P2P navigation to simplify setup processes and visual techniques to forgo infrastructure dependency. Yet, challenges persist: P2P navigation often grapples with path scarcity, and visual techniques on smartphones face issues such as scale uncertainty, directionality limitations, and high computational demands. Addressing these hurdles, our innovation introduces a visual P2P navigation system enhanced by an extensive crowdsourcing framework, effectively mitigating these issues. Through theoretical insights and empirical evidence, our approach demonstrates the potential of P2P navigation—historically constrained by limited path availability—to flourish with an enriched crowdsourced map. We leverage IMU data to overcome scale discrepancies and develop a novel navigational approach to counteract the issue of directionality. Our proposed mobile-edge computing framework facilitates swift navigation (30 FPS with a 100 ms end-to-end latency) while significantly reducing smartphone workload (extending battery life by 35% for a continuous 2 h and 35 min of use), ensuring both precise localization and map accuracy. Our testing confirms a navigation success rate of 100% and a spatial deviation less than 3.2 meters, surpassing previous methodologies in performance.