A High-Stability and Resource-Efficient Centroid Extraction Method for Star Maps

To address the challenges of inadequate stability and excessive resource consumption in stellar centroid extraction for spaceborne star trackers, this paper proposes a single-scan centroid extraction method. The core innovation lies in the integration of an improved Run-Length Encoding (RLE) method with a Dynamic Union-Find (DUF) mechanism, which enables the concurrent execution of connected component labeling and equivalence merging during a single raster scan, significantly minimizing memory access frequency and processing latency. Optimized specifically for stellar imaging characteristics, the method maintains sub-pixel-level accuracy while adapting to irregular star morphologies and complex noise distributions. Experimental results demonstrate that for a 2048×2048 star image, the method achieves a centroid output latency of approximately 10 μs and a per-frame processing time of 20.97 ms. The standard deviation of centroid error is ≤0.1 pixels, and the false detection rate is below 1.5% under high-noise conditions. This work provides a highly resource-efficient FPGA solution for low-latency attitude determination in aerospace applications.