@zppz The second part uses a rotary stage to move the Kinect at a constant angular rate, but the odometry information from the servo is not used as an initial estimate for the ICP.
@ccnyroboticslab i see, I figured so. I am doing something similar but so far my feature descriptors are only from depth info - simple planes with a size and location. They're quite steady across consecutive frames but sometimes there are not enough of them to deduce camera movement. Using the accelerometer 'true-up' as a pseudo-plane helps, but I may yet need to use color data... dammit. Either that or force the operator to find at least two good planes.
This is awesome work. Is there anyway we can access the source code?
vnngoc1506 6 months ago
Comment removed
anaf88 6 months ago
this is awesome, I need to try it
lefantome89 9 months ago
The second part does not look like freehand motion...?
zppz 11 months ago
@zppz The second part uses a rotary stage to move the Kinect at a constant angular rate, but the odometry information from the servo is not used as an initial estimate for the ICP.
ccnyroboticslab 11 months ago
@ccnyroboticslab i see, I figured so. I am doing something similar but so far my feature descriptors are only from depth info - simple planes with a size and location. They're quite steady across consecutive frames but sometimes there are not enough of them to deduce camera movement. Using the accelerometer 'true-up' as a pseudo-plane helps, but I may yet need to use color data... dammit. Either that or force the operator to find at least two good planes.
Are you using sift on the rgb?
zppz 11 months ago
Nice. What code is this using?
motters2001 11 months ago
@motters2001 Thanks - I updated the description
ccnyroboticslab 11 months ago