As i mentioned in last posting, due to the change in method of measuring depth, there exists several differences between Kinect v1's depth image with that of Kinect v2.
Usually, there were large amounts of holes in Kinect v1's depth image, including random holes and occlusions.
Left image of above figure shows depth image of Kinect v1. As you can see, occlusions and random holes(white) are very thick. Comparing to this image, right one is depth image of new version, which captures the same scene with Kinect v1's. And we can figure out that holes are mainly exists along the object boundaries, with the small amount. This is the first major difference between two depth image.
There is another noticeable difference between two. As in above right image, on the corner, there exists large blobs of random holes which disappears in Kinect v1's depth image. This is because intensity values of of returned infrared light are week. But these random holes exists corner area of depth image, not the region of interest.
So, Kinect v2's hole-filling problem is not an important issue, because the quality of depth image is improved significantly.
This is the conclusion of my recent work on Kinect v2's depth image. Any opinions of yours are welcome, and let us discuss about it.
Thanks,
Song, Wanbin
M.S. Candidate
Laboratory of Digital Image Processing
Department of Electronic and Electrical Engineering
Dongguk Univ. Seoul, Korea
Homepage : http://dip.dongguk.edu
------------------------------------------------------------------
No comments:
Post a Comment