Linking event-based vision sensor pixel design, biasing, and performance characteristics to application specific requirements (Conference Presentation)

This talk outlines recent advancements in understanding Event Based Sensor (EBS) pixel architecture, characterization, and biasing, emphasizing how deeper analysis in these realms reveals key techniques that can be leveraged to optimize performance for diverse applications with unique requirements. Novel bias techniques for reducing noise event rates are developed based on a statistical analysis of empirical noise measurements and explained through a fundamental discovery about EBS pixel design. These bias techniques are then applied to a specific use case of space domain awareness, resulting in a 2.6x improvement in sensitivity. A physically realistic pixel model is discussed and shown to match pixel response for lightning detection. This pixel model is then used to infer low level parameters for describing event pixel performance, including dark current, change detection thresholds, and temporal response bandwidth. Finally, noise rates for night sky observations are drastically reduced by using a novel proposed method of background rate subtraction. The principles discussed inform future integration of EBS to diverse applications.