光学 精密工程, 2016, 24 (4): 865, 网络出版: 2016-06-06
复杂场景中航天器靶标的快速识别
Fast recognition of aircraft target in complex scenes
航天器靶标 目标识别 图像处理 圆检测 线段检测 边缘跟踪 aircraft target target recognition image processing circle detection line segment detection edge tracking
摘要
基于靶标的几何特征提出了一种快速识别算法来解决目前航天器靶标识别存在的问题。设计了一款带有线段与圆环图案的合作靶标。利用高斯滤波去除图像中的噪声, 运用Canny检测算子得到边缘图像, 并跟踪得到单像素边缘序列。然后, 通过判断非共线四点是否共圆排除大部分不可能构成圆的边缘, 利用同一圆周的两段子弧对应相同圆心和半径的几何特征实现圆检测。最后根据靶标圆与线段的几何关系排除干扰, 实现靶标的准确识别。实验结果表明, 该靶标识别算法对噪声、光照、旋转等不敏感, 能够在多种复杂场景中快速、准确地识别靶标, 处理时间小于125 ms, 满足实时位姿测量8帧的需要。目前, 该算法已经成功应用于工程样机。
Abstract
A fast recognition algorithm based on the geometric characteristics of a target was proposed to pursue the fast and accurate recognition for the spacecraft target. In the algorithm, a target incorporated with linear and circular patterns was designed to facilitate the accurate recognition. The noises in the images were reduced by mean of a Gaussian filter. The object edges were detected by Canny operator and the sequences of edges were obtained by the single pixel tracking method. Then, the non-circular edges were excluded according to whether the four non-collinear points were located on the same circle or not and the circle detection was implemented by the geometrical property that two different arcs on the same circle have the same center and radius. Finally, the interruption was eliminated by the geometric relation between target circle and line, and the target was recognized precisely. The experimental results demonstrate that the algorithm is insensitive to noise, illumination and target rotation and efficiently recognizes the target in the complex scenes. Moreover, the processing time is less than 125 ms, which meets the requirement of real-time pose measurement of 8 frames. The algorithm has been used in an engineering prototype successfully, now.
初广丽, 王延杰, 邸男, 刘艳滢, 金明河. 复杂场景中航天器靶标的快速识别[J]. 光学 精密工程, 2016, 24(4): 865. CHU Guang-li, WANG Yan-jie, DI Nan, LIU Yan-ying, JIN Ming-he. Fast recognition of aircraft target in complex scenes[J]. Optics and Precision Engineering, 2016, 24(4): 865.