Files
Abstract
In the majority of satellite missions, data is collected onboard and processed on the ground. This bottlenecks the amount of scientific data that can be downlinked without compression. However, as computing systems have progressed, missions have begun using high-performance computers onboard for in-situ computation, reducing the need to downlink raw data. The Multiview Onboard Computational Imager (MOCI) will house its computer vision pipeline onboard an NVIDIA Jetson TX2i GPU in order to create digital elevation models from and recognize objects within high-resolution imagery, while only needing to downlink the final data products. For reliable mission success in a constrained satellite body and unforgiving environment in space, measures are taken to ensure both that the underlying system architecture is fault-tolerant in Low Earth Orbit and that the higher level algorithms can operate within architectural limits and agnostic of sensor noise.