First, the main content
issues feature point method
Extraction and construction descriptors of the key points is very time-consuming.
Ignore the information other than the characteristic points.
Where there is no repeat of the texture or texture, difficult to match exactly.
Solutions
optical flow method. Retain key point, but not constructed descriptor. By tracking the optical flow method to circumvent the key point matching and constructed descriptor overhead. After still use pnp, ICP methods.
Calculating key position in the next frame using the direct method. This method still feature points, but like optical flow method, and matching descriptor avoided.
The direct method using gradation information. This way no critical point and descriptor.
Characteristics of the direct method
by only pixel grayscale camera motion estimation, feature points is not configured.
Photometric optimized by minimizing the error, rather than minimizing the reprojection error.
You can build dense map.
It requires strong assumptions: gray invariance, assumptions window.
Avoid the situation wherein the calculation time and missing features.
Second, the optical flow
optical flow method described pixel motion in the image. Into sparse and dense optical flow optical flow. LK method presented here is representative of a sparse optical flow. The same method is assumed to gradation through a certain pixel transform
Third, the direct method
Because of the time, the above reproduced in part, after October into the original.
----------------
Disclaimer: This article is the original article CSDN bloggers "KylinQAQ", and follow CC 4.0 BY-SA copyright agreement, reproduced, please attach the original source link and this statement. .
Original link: https: //blog.csdn.net/pikachu_777/article/details/83686661