Abstract: To guarantee the celerity and accuracy of identification,an improved algorithm of fundamental matrix estimation was proposed for image matching.Several algorithms were integrated to obtain more accurate match points,and sub-pixel Harris corner detection was applied to extract match point.Based on difference sum theorem,the normalized cross-correlation method was adopted to make coarse matching,and the fast clustering method was used to filter match points.After that,a random sampling method in Statistics was introduced to optimize the matching points and improve Hartley 8 point's method,which was a common method to estimate fundamental matrix.By taking the standard image as the object and the average pole distance as the evaluation index,the proposed algorithm was compared with other commonly used methods.The matching and positioning experiments for an actual target verified the effectiveness of the proposed algorithm in matching speed and accuracy of positioning.

Key words: robot vision, target recognition, image matching, fundamental matrix estimation algorithm

摘要： 为实现智能制造中机器人搬运时的目标物体快速准确识别,提出了一种面向图像匹配的基础矩阵估计改进算法。为获取准确的匹配点,精选并集成了多种方法,并应用亚像素级Harris角点检测方法进行匹配点提取,基于差分求和定理改进的归一化互相关算法进行粗匹配,并采用快速聚类法进行匹配点提纯。在此基础上,通过引入统计学中的分块随机抽样法对匹配点进行优选,实现对常用的基础矩阵估计算法——Hartley8点法的改进。并以标准图像为实验对象,平均对极距离为评价指标对所提算法与其他常用方法进行了比较。最后,进行了实际目标的图像匹配和定位实验,验证了所提算法在匹配速度与匹配精度方面的有效性。

关键词: 机器人视觉, 目标识别, 图像匹配, 基础矩阵估计算法