ravijo
commented
3 years ago @jyp0802
Thank you so much for reporting issues and contributing to improve ros_openpose.
I remember that the depth encoding is different in the case of azure kinect. The non-synchronous version, which was implemented in C++ keeps a note of this information. Please see the highlighted code here.
As you can notice that in the case of '16UC1' encoding, the depth values are represented in millimeters. Therefore a manual conversion from millimeter to meters is required.
In short, you have to implement a similar conditional statement in the scripts/ros_openpose_synchronous.py file.
Please let me know if you need more help. Furthermore, if you can incorporate the workaround, I kindly request you to make a pull request so that it can be merged here.
Finally, I am tagging contributors @QuantuMope and @luorui93 to keep them informed about the possible changes!
jyp0802
First of all, thanks for your work.
When launching
run.launchwithsynchronous:=trueandcamera:=azurekinect, the skeleton markers doesn't show on the point clouds in RViz. Instead if you look carefully at the origin (0,0) you can see markers all together. It's probably because the depth values of azurekinect are at a different scale than for realsense.So when I changed the lines 150, 154, 158 in
ros_openpose_synchronous.pyto multiply 1000, it worked fine.*Like this -> `b_XYZ = self.compute_3D_vectorized(pose_kp, depth) 1000`**
But this is only for the azurekinect. For the realsense camera, it needs to be without the
* 1000. (I haven't tried the other cameras though)Also, when I don't use
ros_openpose_synchronous.py(i.e.synchronous:=false) then it works fine without any changes.