An Evaluation of Low-Cost Vision Processors for Efficient Star Identification

Area of Science:

• Aerospace Engineering
• Computer Vision
• Embedded Systems

Background:

• Star trackers are crucial for satellite attitude determination, demanding high accuracy and low power consumption, especially for small satellites.
• Traditional star identification methods using lookup tables are being superseded by neural network approaches for enhanced performance.
• The Intel Movidius Myriad 2 Vision Processing Unit (VPU) and STM32 Microcontroller were selected for their cost-effectiveness and suitability for computer vision tasks.

Discussion:

• This study compares the accuracy and power efficiency of the Myriad 2 VPU and STM32 Microcontroller for running star identification neural networks.
• The evaluation focuses on the practical application of these processors in low-power, high-accuracy star tracker systems for spacecraft.
• Experimental results provide a direct comparison of performance metrics relevant to embedded navigation systems.

Key Insights:

• The Intel Movidius Myriad 2 VPU achieved 99.08% accuracy with approximately 1 Watt power consumption, even with false stars in the input.
• The STM32 Microcontroller delivered comparable accuracy of 99.07% with similar low power measurements.
• Both processors present viable, cost-effective solutions for implementing advanced star tracking capabilities in resource-constrained small satellite missions.

Outlook:

• Further research could explore optimizing neural network models for these specific processors to potentially improve performance further.
• The findings support the integration of advanced computer vision algorithms into low-power embedded systems for future space missions.
• This comparative analysis provides valuable data for mission designers selecting processors for next-generation small satellite navigation systems.