天基序列星图高精度在轨实时配准方法

High-precision in-orbit real-time registration method for space-based star maps sequence

针对天基序列星图配准存在的效率和精度问题,提出了一种天基序列星图高精度在轨实时配准方法。对基准图和待配准图进行分块,通过预设搜索框的方式,采用归一化相关系数匹配法,获得整像素精度的子图匹配结果;对该结果的小邻域进行二次曲面拟合,获得亚像素精度的匹配对点;使用单应变换模型求解星图的变换关系,完成单帧星图的配准;使用基于末帧的窗口滑动方式,实现序列星图的连续实时配准。实验结果表明,所提方法可以避免恒星质心提取误差和天基平台长时间运动对配准精度造成的影响,相比标准的块匹配,对于16位4096×4096的真实天基星图,所提方法的子图匹配计算效率提升98.7%,配准时间缩短了89.0%,单帧配准的峰值信噪比平均提升了将近0.20;所提序列星图的连续配准方式,速度达到了0.040秒/帧,配准结果的峰值信噪比为6.765,能满足高精度和高实时性要求,通过适当移植优化有望应用于后续的天基平台的弱小目标检测。

Aiming at the efficiency and accuracy issues in registering space-based star image sequences,a high-precision real-time registration method for space-based star image sequences is proposed.Firstly,the base image and the image to be registered are divided into blocks.By presetting search boxes,the normalized correlation coefficient matching method is used to obtain matching results of sub-images with whole-pixel accuracy.Then,a quadratic surface is fitted to the small neighborhood of the result,and matching point pairs with sub-pixel accuracy are obtained.After that,a homography model is used to solve the transformation relationship of the star images and complete the registration of a single frame of star images.Finally,a window sliding method based on the last frame is used to achieve continuous real-time registration of the star image sequences.Experimental results show that the proposed method can avoid the influence of errors in extracting star centroids and the long-term motion of space-based platforms on registration accuracy.Compared with the standard block matching method,for a real 16-bit space-based star image with a size of 4096×4096,the proposed method improves the computational efficiency of sub-image matching by 98.7% and reduces the registration time by 89.0%.The average peak SNR of single frame registration is improved by nearly 0.20.Meanwhile,the speed of the proposed continuous registration method of star image sequences reaches 0.040 seconds per frame,and the peak SNR of the registration results is 6.765,which can meet the requirements of high accuracy and high real-time performance.Through appropriate transplantation and optimization,this method is expected to be applied to the detection of weak and small targets on subsequent space-based platforms.