基于多尺度分析的快速相位立体匹配

Fast Phase-based Stereo Matching Based on Multi-scale Analysis

摘要:

基于相位的立体匹配是双目投影光栅相位法中的重要步骤,但传统的相位匹配方法在处理高分辨率图像时因存储空间大大增加,难以达到匹配速度与精度的平衡。文章提出了一种基于多尺度分析的快速相位立体匹配算法,采用分层匹配的策略,对预处理后的左右绝对相位图进行降采样以生成图像金字塔,利用低分辨率的视差匹配结果以预测下一层视差,以此降低下层高分辨率图像的视差搜索范围,达到匹配速度与精度的平衡。实验结果表明,所提算法在保证精度的情况下能有效提升相位立体匹配速度,实现高分辨率相位图快速准确的立体匹配。

Abstract:

Phase-based stereo matching (PSM) is a vital step in binocular structured light measurement, but the traditional PSM method is difficult to strike a balance between speed and accuracy due to the greatly increased storage space when processing high-resolution images. In this paper, an image pyramid model is introduced and a phase-based stereo matching algorithm based on multi-scale analysis is proposed. With a layered matching strategy, the preprocessed left and right absolute phase maps are down-sampled to generate an image pyramid and high-resolution disparity maps are predicted by up-sampling low-resolution disparity maps to narrow the disparity range. Experimental results indicate that the proposed algorithm can greatly improve the phase-based stereo matching speed while ensuring the accuracy and can achieve fast and accurate stereo matching of high-resolution images.