Closed HJMGARMIN closed 7 months ago
Hi,
Regarding this part of the description, does it mean that for a 3D line in the same space, its endpoints may not be exactly the same when the corresponding line segments in different observation frames are back-projected into the 3D space?
Yes. In the paper (Sect. III.C), there are a couple of sentences that motivate this point: Due to noise, possible occlusions, and other factors, the 3D line segment l may not be fully observed by the camera and endpoints are often inconsistent across images [8]. Furthermore, a 3D line segment can be fragmented into multiple line segments (belonging to the same line in 3D).
Then, this article also mentions P and Q when representing 3D lines. Is there a specific code description for PQ in this picture?
Yes. The MapLine
class hosts a representation of the two 3D points P
(start) and Q
(end) of the 3D line segment (in world coordinates):
Eigen::Vector3f mWorldPosStart;
Eigen::Vector3f mWorldPosEnd;
The 2D image observations of the MapLine
are the 2D keylines cv::line_descriptor_c::KeyLine
, each one hosting the image coordinates of the start and end points (pixel coordinates).
Hope this helps.
For 2D lines matched in two frames, they are back-projected to the same line segment in three-dimensional space. If their back-projected space endpoints do not coincide, will P and Q make the two back-projected line segments coincide?
Regarding this part of the description, does it mean that for a 3D line in the same space, its endpoints may not be exactly the same when the corresponding line segments in different observation frames are back-projected into the 3D space?Then, this article also mentions P and Q when representing 3D lines. Is there a specific code description for PQ in this picture?![image](https://github.com/luigifreda/plvs/assets/52454672/396fda21-78e6-4bc0-85e0-f7bd8e53ac87)