Abstract
Event cameras are bio-inspired sensors that offer several advantages, such aslow latency, high-speed and high dynamic range, to tackle challenging scenariosin computer vision. This paper presents a solution to the problem of 3Dreconstruction from data captured by a stereo event-camera rig moving in astatic scene, such as in the context of stereo Simultaneous Localization andMapping. The proposed method consists of the optimization of an energy functiondesigned to exploit small-baseline spatio-temporal consistency of eventstriggered across both stereo image planes. To improve the density of thereconstruction and to reduce the uncertainty of the estimation, a probabilisticdepth-fusion strategy is also developed. The resulting method has no specialrequirements on either the motion of the stereo event-camera rig or on priorknowledge about the scene. Experiments demonstrate our method can deal withboth texture-rich scenes as well as sparse scenes, outperformingstate-of-the-art stereo methods based on event data image representations.