An adaptive spot-centroid extraction method is proposed to address low extraction accuracy and insufficient real-time performance in cooperative-target feature-point imaging for large-span dynamic measurement， problems that arise from pronounced size variation and edge jitter. Exploiting the motion coherence of cooperative targets， a dynamic ROI feature-parameter model is first established to enable rapid and accurate ROI localization via interframe motion prediction， thereby substantially reducing the data volume for subsequent processing. Canny edge-detection parameters are then adaptively adjusted using an Otsu-based threshold optimization strategy， which improves noise suppression across varying measurement distances while enhancing computational efficiency. Sub-pixel edge localization is refined by combining a multi-directional Sobel operator with Zernike moments enhanced by an intensity ramp， and centroid coordinates are obtained via a robust least-squares circle fitting method improved through Gauss-Newton iteration. Validation on simulated image datasets and experimental measurements demonstrates that， for small-scale spots with simulated edge blur， the centroid positioning error of the proposed method ranges from 0.001 to 0.025 pixels under different noise levels. In practical tests， the ROI prediction algorithm satisfies measurement scenarios with accelerations up to 8.75 m/s²， and the repeatability error of spot positioning at measurement distances of 10-30 m remains stable between 0.016 and 0.040 pixels， outperforming conventional methods. Meanwhile， spot-extraction speed is increased by approximately 75.5%， markedly improving real-time processing capability. The proposed approach offers effective technical support for cooperative-target measurement applications.

cooperative target detection;adaptive spot centroid extraction;inter-frame motion prediction;improved Sobel-Zernike moment