An FPGA-SoC Camera Payload Processing Unit with Soft-GPU Acceleration for Onboard Vision

This paper presents an FPGA-SoC-based camera payload processing unit for onboard vision on resource-constrained spacecraft platforms. The system combines Engineering Minds Munich’s Smart Power and Processing Module (SPPM), a remote camera interface based on MIPI-CSI-2, and a GPU-accelerated embedded processing chain using a soft-GPU concept. The architecture is intended for image-based monitoring and recognition tasks directly on the payload electronics, enabling local interpretation of visual data instead of relying exclusively on transmission of full-frame imagery. A lightweight semantic-segmentation network identifies relevant scene structures, while deterministic post-processing kernels convert the dense output into compact geometric descriptors and other image-derived outputs for subsequent interpretation. In the current engineering implementation, the processing chain is distributed between the FPGA-based camera payload platform and an external GPU@SAT accelerator used as a prototype soft-GPU environment. Experimental validation in an EGSE environment demonstrates the feasibility of the end-to-end processing chain and highlights current challenges related to illumination variability, background complexity, memory-aware implementation, and execution time. The presented approach shows how soft-GPU acceleration can extend FPGA-based payload electronics toward onboard vision with compact, decision-relevant outputs.