Open fraukej opened 4 years ago
Photometric error is being used in stereo matching. In order to compute/update the depth of a pixel, they first compute the 5-equidistant points (think of it as a 'descriptor') in the current keyframe, and then compute the corresponding 5-equidistant points along the epipolar line in the reference (or current) frame. A correspondence is considered found where the re-projection location has the minimum photometric error.
In order to compute the photometric error, you need the relative transformation between the current keyframe and the current frame. To estimate depth, you just simply solve the unknown d given the projection matrix and the correspondence.
Can somebody explain to me what the photometric error is? I can't find an intuitive exmplanation.
I understand that it's the error of pixel intensity values of the same pixel in two frames. But how to I know the homography between the two pixels if I don't know the relative pose? And how does that help me to estimate depth?