Rachel Oliver, Air Force Institute of Technology; Dmitry Savransky, Cornell University
Keywords: Event-based Vision Sensors, Neuromorphic Cameras, Space-based Sensing, Space Domain Awareness, Noise Modeling
Abstract:
Building off the foundation of a physics-based end-to-end model for event-based vision sensors (EVS) observing resident space objects (RSOs), we apply new techniques to model realistic low-light sensor noise. While previous event-generation approaches simulate memorized current leakage and apply temporal noise models, our methods improve on these approaches and additionally account for current-following white noise as an event source. These model improvements are key components towards accurate event-generating simulations which can advise requirements and concepts of operations for dedicated event-based Space Domain Awareness (SDA) architectures. Simulating event-generation that models the underlying physics is particularly important for space-based architectures where informative space-based truth data to advise design is limited. The EVS’s pixel’s independent and asynchronous recording of changes in photocurrent produces data with high temporal resolution and dynamic range making it an attractive technology for SDA. EVS are particularly appealing as a space-based payload because of their sparse data output reducing the need for downlink, computational, and power resources. To enable creation of a space-based EVS system, our physics-based end-to-end model now includes noise based on induced photocurrent to generate simulated events closer to known truth. We introduce a new Poisson-based method to model the noise generated by the temporal variation of the dark current and a method to tune high-frequency white noise on the induced photocurrent to model the noise on signals above the dark current. These techniques demonstrably improve EVS noise modeling by closely matching known truth event rate and polarity behavior, moving EVS one step closer to operational space-based SDA usage.
Date of Conference: September 17-20, 2024
Track: Space-Based Assets