Please use this identifier to cite or link to this item:
Title: Acceleration of Computer Vision Algorithms for Star Trackers on SoC FPGA Platforms
Authors: Πανουσόπουλος, Βασίλειος
Σούντρης Δημήτριος
Keywords: Star Tracker
Star Centroiding
Hardware Optimization
Center of Gravity
Fast Gaussian Fitting
Issue Date: 13-May-2022
Abstract: Space applications demand precise and fast measurement of the satellite’s orientation, which can only be achieved with the use of star trackers. This instrument consists of a camera sensor which captures sky images and appropriate digital hardware that detects the stars within the image and matches them to known maps, targeting to determine the satellite’s attitude in the inertial space. The large amount of sensor data makes the task of detecting stars computationally intensive and therefore its execution on a general purpose embedded processor is inefficient. Additionally, the latest trends of utilizing Commercial Off-The-Shelf (COTS) HW in space leads us to examine such components in the processing architectures, i.e., a high-resolution camera and a high-performance COTS FPGA. In this thesis, we present the acceleration of the centroiding process on COTS FPGA platforms, during which the precise position of each star in the image is estimated. Two centroiding algorithms which demonstrate different levels of accuracy and complexity have been adopted. We optimize and implement these algorithms on hardware using both VHDL and HLS (C++) as FPGA programming methods and perform extensive design space exploration to find the most efficient solution. An in-depth study of the advantages that each method provides is also presented. The hardware models are validated using simulated star images and evaluated in terms of computation time and centroiding accuracy. The simplicity of the Center of Gravity algorithm enables a high degree of acceleration by 2 orders of magnitude but its accuracy might be considered insufficient. Therefore, we propose a fully pipelined novel FPGA implementation that features excellent accuracy and high efficiency. This design adopts the Fast Gaussian Fitting algorithm which is accelerated by 25x. A thorough analysis that describes the limitation of achieved acceleration due to high complexity and high FPGA resource utilization is provided. Ultimately, this architecture is suitable for real-time applications as it removes the bottleneck of data processing and can enable parallel execution of the processes involved in a star tracker.
Appears in Collections:Διπλωματικές Εργασίες - Theses

Files in This Item:
File Description SizeFormat 
DT_Panousopoulos_Vasileios.pdf4.21 MBAdobe PDFView/Open

Items in Artemis are protected by copyright, with all rights reserved, unless otherwise indicated.