【摘要】 针对传统Census立体匹配算法在弱纹理和边缘区域匹配精度较差的问题，提出一种基于特征信息优化的代价计算方法，在窗口中融入更多的差异信息以获得更精确的像素视差值。随后采用多方向路径独立的线扫描优化计算聚合代价以进一步提高匹配精度。为获得更好的遮挡区域匹配效果，提出一种基于差异填充的视差优化方法，对遮挡像素进行识别和视差填充。为提高算法的效率，提出一种基于降采样策略的算法运行模式，通过缩小视差搜索范围以减少硬件负荷。最后以五组标准图像为输入进行改进Census算法性能检验，结果显示，平均误匹配率为6.12%,较改进前降低了2.45%,算法效率平均提升17.7%。更多还原

【Abstract】 Aiming at the problem of poor matching accuracy of traditional Census stereo matching algorithm in weak texture and edge areas, we propose a cost calculation method based on feature information optimization, which integrates more difference information into the window to obtain more accurate pixel disparity value. Subsequently, multidirectional path independent line scan optimization was used to calculate the aggregation cost to further improve the matching accuracy. In order to obtain better occlusion area matching effect, a disparity optimization method based on difference filling is proposed to identify the occlusion pixels and make disparity filling. In order to improve the efficiency of the algorithm, a new algorithm operation mode based on the downsampling strategy is proposed to reduce the hardware load by narrowing the disparity search range. Finally, the performance test of the improved Census algorithm was conducted with five sets of standard images as input. The results showed that the average mismatching rate was 6.12%, which was 2.45% lower than before the improvement, and the average efficiency of the algorithm increased by 17.7%.更多还原