Environment Understanding with EdgeSLAM

Accurate and efficient instance segmentation on mobile platforms is crucial for applications in augmented reality, inspection tasks, and understanding environments. Despite the development of edge-assisted systems to facilitate computation-intensive operations, the unique challenges of instance segmentation, including device and target motion, precision requirements, and computational burdens, limit its performance on resource-constrained devices. This chapter introduces edgeIS, an innovative system combining mobile and edge computing resources to deliver real-time, precise instance segmentation. edgeIS leverages mobile device capabilities to sense environmental dynamics and its motion, fostering a novel collaboration between mobile devices and edge servers specifically tailored for segmentation tasks. Implemented across various mobile devices and a lightweight edge server, edgeIS undergoes rigorous testing across three datasets, demonstrating its capability to operate in real-time with a segmentation IoU of 0.92, surpassing current leading solutions. Moreover, edgeIS’s integration into an augmented reality inspection system within an oil-field showcases its industrial applicability and performance excellence.