Event-Based Spacecraft Representation Using Inter-Event-Interval Adaptive Time Surfaces

This paper introduces the Inter-Event-Interval-Adaptive Time Surface, a novel event-based algorithm for reconstruction of event-frames under highly dynamic object motion and a range of lighting conditions. Event cameras are of particular interest in the proximity operations sector as they offer microsecond latency, 120 dB of dynamic range, sparse data output, and low power requirements. These benefits make event cameras highly suitable for handling scenes where traditional cameras struggle: extreme lighting conditions and dynamic motion. The IEI-ATS algorithm utilizes a spatially-varying decay mechanism where every pixel's exponential decay constant $\tau(x,y)$ is dynamically adjusted based on the local temporal statistics derived from inter-event-intervals. The proposed approach enhances edge preservation while minimizing motion blur without any optical flow calculations. The algorithm's decay mechanism naturally suppresses static background elements, making it particularly effective for stationary observation of moving targets. Experimental validation of the algorithm using Carleton University's Spacecraft Proximity Operations Testbed is used to demonstrate the effectiveness of the approach when observing a rotating and translating target under (1) nominal lighting conditions, (2) low-light conditions, and (3) harsh lighting conditions. A comprehensive public dataset (EVOS: EVent-based Observation of Spacecraft) was also created to support the validation of this algorithm; the dataset includes a diverse set of spacecraft maneuvers under various lighting conditions, with corresponding ground truth data for the position and velocities of a chaser and target spacecraft. Overall, the results confirmed that the IEI-ATS algorithm was suitable for reconstructing time surface representations of tumbling spacecraft during autonomous on-orbit inspection tasks, and proved particularly effective in dark and bright scenes where traditional optical sensors are ineffective.