立体匹配是计算机视觉中一个重要的研究课题。目前主要的全局匹配方法虽然能得到比较准确的立体匹配结果,但是其计算速度较慢。通过对一种自适应分层置信传播算法的分析,对其进行了并行实现,加快算法的计算速度。该立体匹配方法在传统的HBP算法基础上对消息计算的收敛性进行判断,停止消息计算已收敛节点的迭代计算,从而减少算法整体迭代次数。通过利用该算法迭代过程中的并行性,对该算法进行并行实现。实验表明,这里并行实现方法在不损失计算效率的前提下,在双核处理器上平均加速比达到1.88。
stereo matching is an important research topic in the field of computer vision. Although the global matching method could achieve more accurate stereo matching results, its calculation is fairly slow. A HBP algorithm is studied, and its time-consuming part paralleled, thus to hasten its operation speed. This method could reduce computation quantity by stopping the iterations of messages, which are already converged to optimal values. The experimental results show that this method has little influence on the calculation accuracy, and could achieve an average speed of 1.88 on a dual-core processor.