Event cameras are bio-inspired, asynchronous sensors that detect changes in pixel brightness, providing data updates at a microsecond temporal resolution with low power consumption. Due to their increased dynamic range and spatiotemporal sensitivity, these cameras can capture dim objects that are moving very fast. These benefits make this type of camera well-suited to space applications. Despite these advantages, event cameras introduce a new paradigm for computer vision algorithms, with updates occurring on a pixel-by-pixel basis rather than a traditional frame-by-frame approach, rendering most existing perception methods unsuitable. The constrained onboard resources of satellites further challenge the development of suitable algorithms. Event cameras are highly prone to noise, particularly under low-light and space environmental conditions, making robust onboard filtering algorithms essential to reduce the volume of data to be downlinked and to prevent true signals from being obscured by noise.
This paper presents the experimental characterization of noise in event-based cameras for a CubeSat technology demonstration mission and discusses preliminary approaches to modeling its temperature dependent behavior. We first describe our thermal chamber test setup and the procedures developed to evaluate the sensor under low light conditions representative of space. Using a thermal chamber, we recorded multiple scenes across temperatures points ranging from −20°C to +60°C. Then, we present our noise sensitivity results from thermal chamber testing. We show that hot pixels dominate the event stream, contributing 94.5–98.8% of events depending on scene and temperature, and occur most frequently at low temperatures, disappearing entirely at +44°C. We discuss different hot pixel removal strategies with attention to mitigating false positives. Analysis of the remaining signal shows a polarity imbalance with negative events dominating at low temperatures and varying retention rates across temperatures, highlighting temperature-dependent behavior in the background activity. These results provide first guidance for temperature dependent calibration and in orbit noise filtering strategies. We then discuss the limitations of experimental noise modelling and present key lessons learned from our experimental test.
«
Event cameras are bio-inspired, asynchronous sensors that detect changes in pixel brightness, providing data updates at a microsecond temporal resolution with low power consumption. Due to their increased dynamic range and spatiotemporal sensitivity, these cameras can capture dim objects that are moving very fast. These benefits make this type of camera well-suited to space applications. Despite these advantages, event cameras introduce a new paradigm for computer vision algorithms, with updates...
»
Login
BibTeX