16 double values for one joint

[jiangwei221]

16 double values for one joint

[ashafaei]

Hello @jiangwei221,
The representation of posture is not just the XYZ. You could also characterize the rotation of the joint with a 3x3 rotation matrix. The 16 values are actually a flattened 4x4 matrix. If you reshape it to 4x4 you'll get a 3x3 rotation matrix on the left top corner, and a 3x1 offset vector on the right side (those are the 13:15 values).

   -4.8039   -1.1965   -8.6886  -11.1906
    1.5543    9.6336   -2.1860   45.5783
    8.6317   -2.4006   -4.4419  -23.3961
         0         0         0    1.0000

This representation of joint is more informative than the mere xyz values, but it's rarely inferred in computer vision applications.

Yes, you are right. The joint locations are in the Maya space, whereas, the reconstruction is in the Kinect space. You need to use the extrinsic camera parameters (in the groundtruth file for each camera) to properly rotate and translate the point cloud to merge them. This is also implemented in the code process/generate_cloud_camera.m.

[jiangwei221]

Thank you for the reply! I understand now!