Stereo Matching (Kitti Benchmark)

Kitti的stereo matching benchmark有2012和2015两个版本。

在2012的版本中，ground truth是利用ICP从先后五帧的点云中求得相对pose，和积累得到的点云，利用camera的参数信息反投回图像，再人为的移除镜面反射的区域得到。

在和其他benchmark，比方说middleburry的stereo数据集的比较过程中发现，一些基于local patch的方法在middleburry上面表现得很好，但是在kitti上面表现的很差。在kitti的论文中分析了这一现象，认为原因是kitti数据集的场景中有很多non-texture area比方说路面，墙面和阴影，区别于middleburry数据集丰富纹理的场景，很难让local的方法得到好的performance

平行平面假设

Our evaluation table ranks all methods according to the number of non-occluded erroneous pixels at the specified disparity / end-point error threshold. All methods providing less than 100 % density have been interpolated using simple background interpolation as explained in the corresponding header file in the development kit. For each method we show:

Out-Noc: Percentage of erroneous pixels in non-occluded areas

Out-All: Percentage of erroneous pixels in total

Avg-Noc: Average disparity / end-point error in non-occluded areas

Avg-All: Average disparity / end-point error in total

Density: Percentage of pixels for which ground truth has been provided by the method

在2015年kitti的另外一篇论文中(Object Scene Flow for Autonomous Vehicles)，提出了新的stereo和optical flow benchmark，里面有更多的含有动态物体的数据