Accelerating AI and Computer Vision for Satellite Pose Estimation on the Intel Myriad X Embedded SoC

TL;DR

This work tackles onboard satellite pose estimation under strict power and radiation constraints by developing a hybrid AI/CV system on the Intel Myriad X VPUs. It combines UrsoNet, a ResNet-50–based DNN, with Lourakis-based CV tracking, and implements extensive low-level optimizations across the NCE/SHAVE platform to achieve up to $5$ FPS on 1-MegaPixel RGB inputs within a $2$ W envelope. The study provides a detailed evaluation against embedded CPUs/GPUs and Myriad predecessors, showing substantial speedups and favorable power efficiency for both the AI and CV components, while maintaining acceptable pose accuracy. The results demonstrate the practicality of a single-chip onboard solution for Lost-In-Space and satellite tracking with potential applicability to mixed-criticality space avionics and in-flight programmability through the OpenVINO/MOVENCI toolchain. Overall, the paper validates heterogeneous SoCs as viable platforms for autonomous space vision tasks and outlines directions for improving cross-pipeline coordination and mission-specific tuning.

Abstract

The challenging deployment of Artificial Intelligence (AI) and Computer Vision (CV) algorithms at the edge pushes the community of embedded computing to examine heterogeneous System-on-Chips (SoCs). Such novel computing platforms provide increased diversity in interfaces, processors and storage, however, the efficient partitioning and mapping of AI/CV workloads still remains an open issue. In this context, the current paper develops a hybrid AI/CV system on Intel's Movidius Myriad X, which is an heterogeneous Vision Processing Unit (VPU), for initializing and tracking the satellite's pose in space missions. The space industry is among the communities examining alternative computing platforms to comply with the tight constraints of on-board data processing, while it is also striving to adopt functionalities from the AI domain. At algorithmic level, we rely on the ResNet-50-based UrsoNet network along with a custom classical CV pipeline. For efficient acceleration, we exploit the SoC's neural compute engine and 16 vector processors by combining multiple parallelization and low-level optimization techniques. The proposed single-chip, robust-estimation, and real-time solution delivers a throughput of up to 5 FPS for 1-MegaPixel RGB images within a limited power envelope of 2W.

Figures (6)

• Figure 1: Intel's Myriad X VPU paralos.
• Figure 2: Cooperation of AI and CV for pose estimation/tracking on Myriad X.
• Figure 3: The architecture of the UrsoNet DNN for pose estimation ursonet.
• Figure 4: Mapping of the DNN for pose estimation on Myriad X.
• Figure 5: The block diagram of the CV pipeline for pose tracking lourakismyriad2.