Boosting event-based structured light via cyclic spatial coding

Event-based structured light (SL) systems leverage a novel neuromorphic sensor, the event camera, to realize high-speed, high dynamic range, and low power consumption depth sensing. In this field, the spatial coding method, e.g., random speckle, offers the highest scanning speed but has moderate reconstruction quality, whereas point or line scanning methods provide higher accuracy but are more time-consuming. To boost accuracy while maintaining efficiency, we design a novel cyclic spatial coding strategy for event-based SL systems, which embeds geometric constraints among multiple consecutive frames through the cyclic displacement of the random speckles. By exploiting the embedded geometric constraint, we further develop a geometric prior-guided stereo matching algorithm to achieve high-accuracy depth reconstruction. Additionally, we propose a multi-frame enhancing strategy to improve the quality of event frames, providing a robust basis for the stereo matching. Experiments demonstrate that the proposed method effectively boosts the performance of event-based SL systems, achieving high-accuracy 3D sensing at 1000 FPS.