摘要
针对移动机器人在曲面场景的匹配问题中,同形约束用于解决极约束产生的匹配模糊性问题和发现新的匹配点。实际上是平面块对曲面进行近似逼近的过程。逼近程度和逼近性能需要有指标进行定性和定量的衡量。故提出了两种性能评价指标:平均映射误差和平均映射匹配对。仿真实验结果的分析证明,场景深度变化或者场景距离摄像机的距离变化,对立体匹配算法性能本身不受影响,但映射和建立匹配关系时所需要的同形矩阵的数量不同。而且,随着特征点的稠密度提高,曲面场景的稳定性降低,可随着迭代过程的进行,算法本身结果还趋于稳定。
Local planar homographies （plane projective transformations） are used to reject outliers produced by epipolar constraint for curved scenes. Moreover, local homographies are used to find new approximate matches by predicting feature points. This process is an approximation using many planes to approximate curved scenes. In order to evaluate the approximation performance, two indexes, average mapping error and average mapping match pair are proposed to analyze the performance of the algorithm in this paper. Experimental results with real image data illustrate the efficiency of the indexes.
出处
《计算机仿真》
CSCD
北大核心
2009年第7期260-263,共4页
Computer Simulation
基金
上海市重点学科建设项目(P1303)
关键词
图像校准
特征匹配
弱标定
同形矩阵
对极约束
Image rectification
Feature correspondence
Weakly calibrated
Homography
Epipolar constraint
