Lack of Documentation, ZED SDK

For ZED SDK for body tracking with Body 38 with the following parameters: "action_state”, "bounding_box_2d”, “confidence”, "dimensions”, “global_root_orientation”, “head_bounding_box”, head_bounding_box_2d”, “head_position”, "id”, keypoint”, “keypoint_2d”, “keypoint_confidence”,” keypoint_cov”, “position”, “position_covariance”, “tracking_state”, "unique_object_id”, “velocity”, "is_new”,” is_tracked”, “timestamp”

  1. There is no documentation on “position_covariance” and “keypoint_cov”.

----Why are there six outputs of “position_covariance” and “keypoint_cov” and what do each of them represent? For each, between what two variables is the covariance measured?

----What are the similarities and differences between “position_covariance” and “keypoint_cov”? My understanding is that “position_covariance” is the six covariance values of some variables within the entire frame while keypoint_cov is some form of covariance between two variables for a specific joint.

The documentation about detection outputs does not describe anything about covariance: Body Tracking Overview | Stereolabs

  1. Please elaborate on the documentation for global root orientation. What is the difference between this and the

  2. Documentation writes that there is keypoint (x,y,z) and local position per joint (x,y,z). However, there is no parameter called “local position per joint”, only a parameter for position called “keypoint.” There is also a parameter called “local orientation per joint” (x,y,z,w). What does this correspond to in the parameters in the json file? And what is the difference between the x,y,z values of “local position per joint” versus “local orientation per joint”. For the w value of local orientation per joint, please elaborate on what w specifically represents.

  3. Do you plan to support angular position for the joints like Azure Kinect DK supports?: Azure Kinect body tracking joints | Microsoft Learn

1] you can find Documentation here:

2] the position_covariance is about the root position whereas the keypoint_cov is for every keypoint

3] the keypoint values are the 3D positions (x,y,z) whereas the keypoints orientation are the orientation quaternion for every keypoints.

4] we provide almost the same data, expect ours are not global orientation but local (respect to the parent keypoint)