Event-based Spinning Object SLAM

Ethan Elms, Yasir Latif, Tat-Jun Chin; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2025, pp. 4655-4664

Event sensors offer high temporal resolution and high dynamic range visual sensing which makes them appealing for visual simultaneous localization and mapping (VSLAM). In this work, we investigate event-based VSLAM for the scenario where a static event camera observes an object that is undergoing an unknown spinning motion. Geometrically, the setting is equivalent to a static object that is observed by an event camera that is orbiting around the object with unknown orbital parameters. We exploit this duality to develop an algorithm called eSpinSLAM. Key components of eSpinSLAM are an online event-only feature detection and tracking mechanism and a continuous-time back-end that can incrementally reconstruct the object and estimate the orbital motion. The problem geometry not only permits a camera state representation with a bounded number of parameters that can support infinite time horizon operation, but also enables effective loop closure detection and drift mitigation via spinning frequency estimation. Results on a real event dataset validate the improved feature tracking, higher reconstruction accuracy and greater throughput of eSpinSLAM over existing event-based 3D vision methods.