I. The results show incorrect mapping which have large offsets. Here are first set of ways to check what is the matter:
- Instead of using the Hand-held camera, using a Web-cam as a substitution (...because of unsure factors caused by hand-held camera, even though already set the length in 12mm)
- Instead of using only 4 points, adding more than 4 points
- Instead of using the morphological operation in order to located the central position of the 4 Red Markers, directly assign position or points (...try to minimize any uncontrolled factors)
- Check the result when do "Kinect2Kinect" (...the location of E/EGN points should be the same (or very close) to the ground truth points)
- Make sure all units are the same, i.e., all in "mm" or in "pixel"
- Check the correction in doing coordinate transformation, and check that using the correct coordinate system
- Check "Depth Map" from Kinect is really capture the depth value (...i.e., not distance between Kinect and object is within the sensing range)
No comments:
Post a Comment