Hardware Implementation of Fast and Robust Star Centroid Extraction With Low Resource Cost

Abstract

Star trackers measure the attitude of a spacecraft by matching the stars captured by the camera and those stored in the onboard database, whose directions are already known. The information (i.e., location and brightness) on the stars in the captured image must be correctly and timely provided for star recognition. This process is called star centroid extraction. The hardware implementation of the star centroid extraction algorithm using parallel and pipelined architecture is a proper solution to ensuring higher accuracy as well as lower time cost. However, some limits restrict the performance of these kinds of algorithms. For example, faint stars, disturbing objects (e.g., the moon, bright planets, and so on), and noise pixels are not valid stars but resume a large amount of resource. Some irregularly shaped star spots may cause the algorithms to obtain inaccurate results. To solve these problems, this paper proposes a star centroid extraction method implemented on field programmable gate arrays (FPGAs) with a dynamic rooted tree architecture. In contrast to the traditional connected domain segmentation method, this method merges the equivalence table in the process of scanning, such that only one scan of the image is needed. Moreover, this method profits from a strict equivalence merging logic and can deal with various irregularly shaped star spots. Experiments are performed both on PC simulations and FPGA platforms, and results show that this method achieves good performance at a very low resource cost.