基于鲁棒估计的捕获段面目标图像处理

Extended Object Image Processing Based on Robust Estimation Algorithm During Capture Stage

深空探测捕获段导航图像为含轮廓面目标图像,目标质心与光心并不重合,传统加权质心方法无法有效提取目标视线。文章通过最小平方中值法对提取的面目标导航图像真实边缘轮廓进行椭圆鲁棒拟合,该方法能够有效估计出目标中心视线,同时能够抑制由于目标表面坑洞等纹理特征引起的误差,对伪边缘、拍摄噪声均不敏感。最后,通过大量仿真与真实图像处理,对算法进行验证。结果表明,用文中方法估计的目标中心投影点坐标精度优于0. 1个像素,视线误差优于1×10-4rad,方法具有极高的鲁棒性,对捕获段图像处理有一定的参考价值。

The navigation image is a contoured extended object image during capture stage in deep space exploration,and the center of mass and optical center of the navigation image don't coincide.The traditional weighted centroid method cannot extract the line of sight of target effectively.In this paper,the elliptical robust fitting of the extracted true edge contour of the navigation image is performed by the least squares median method and it can estimate the line of sight of the target with high accuracy.The error caused by complicate texture characteristics,such as surface holes of target,is suppressed and the whole approach is not sensitive to noise and false edge.Experiments with a large number of simulation images and real images are carried out and the result shows that the extracted center of the ellipse by the proposed method can reach 0.1 pixel precision and the error line of sight is better than 1×10-4 rad.The proposed method has good robustness and effectiveness which.has certain reference value for image processing during capture stage.