In traditional shape context, the calculation complexity was high in traversing the contour sampling points to calculate the histogram similarity and it can not meet the industrial real-time requirements. In order to solve the problem, a fast stereo matching algorithm of workpiece images based on the improved shape context was proposed. In the disparity constraints of stereo image centroid, by using the histogram distribution of shape context, candidate matching points set were obtained and the calculation complexity was greatly reduced. To increase the discrimination between the matching points and non-matching points, the similarity measurement of shape context was weighted by the weighted coefficient. During the matching period, the improved shape context was combined with the histograms of oriented gradients of the corresponding contour feature points in 3 × 3 neighborhood to acquire the initial matching point set. At last, the false matching points were eliminated by random sample consensus algorithm. Theoretical analysis and experimental verification were applied to workpiece images, and the corresponding experimental comparison was presented. The experiment results show that this improved algorithm has higher precision, faster matching speed and high robustness.%为了解决传统形状上下文直方图相似性计算需遍历轮廓采样点、计算复杂性高、不能满足工业实时性要求的问题，提出了一种改进型形状上下文的工件立体匹配方法。改进匹配搜索策略，引入立体图像对质心的视差约束条件，并利用形状上下文直方图分布信息进行初步筛选，获取候选匹配点集，减少后续匹配计算复杂度；为增加匹配点与非匹配点的区分度，对形状上下文相似性度量进行加权处理；匹配时融合对应轮廓采样点3×3邻域的梯度方向直方图特征，与形状上下文构成联合相似性度量；最后采用随机抽样一致性算法剔除误匹配点对。对改进算法进行了理论分析，并应用于工件图像进行实验验证，通过实验给出了相应的实验对比。结果表明，改进的方法具有较高的匹配精度和更快的匹配速度，鲁棒性高。
展开▼
