直线特征立体匹配中的不确定性问题

Uncertainty in matching straight line segments

直线特征断裂和提取不完整、过提取是直线特征提取结果中常见的不确定性问题,是影响直线特征立体匹配结果的关键因素。提出一种以边缘检测结果为依据定量描述这些不确定性的方法,依据已提取直线特征延长线附近边缘点分布,定义直线特征端点定位的概率密度函数,以单条直线特征的不确定性描述为基础,计算断裂直线特征编组假设的可靠性测度,进而计算用于衡量立体图像中任意两条直线特征匹配假设可靠性的核线约束测度和灰度相似性测度。应用真实遥感立体图像,对比已有方法,结果证明了本文方法能够有效提高直线特征匹配结果对于两类直线特征提取不确定性问题的抑制能力。

The fragmentation, incomplete extraction and excessive extraction are three kinds of common uncertainty problems in extracting straight line segments, which are critical to matching straight line segments across different images. Based on the analysis of the process of extracting straight line segments, a method is proposed to describe quantitatively the above two problems by the result of edge detection. With the positions and the gradients in gray of edge points close to the extension of straight line segments, the probability in po- sition of endpoints of each straight line segment is defined. According to the above probability expression, the credibility of grouping different lines is measured and furthermore the measure that any two straight line segments extracted from different images satisfy the epipolar constraint and photometric similarity constraint is computed, with which the strength of each potential straight line segment matching pair is given. The experimental results on real stereo remote sensing images show that the method proposed in this paper is competitive.