Still drift of skeleton and lower fps of 4.0 SDK, ZED2

I found skeleton in both 3.7 and 4.0 sdk are still drifting in skeleton data, especially in z direction. When one man walking straightly directly to camera from 6 meteres. While camera itself is fixed.

if there is anyway to deal with this issue, I am appreciated.

here are some of the most important parameters
windows 10, python 3.9, sdk ver:4.0

i_params.camera_resolution = sl.RESOLUTION.HD1080
i_params.coordinate_units = sl.UNIT.METER
i_params.camera_fps = 30
i_params.coordinate_system = sl.COORDINATE_SYSTEM.RIGHT_HANDED_Y_UP
i_params.depth_mode = sl.DEPTH_MODE.ULTRA

b_param.enable_tracking = True
b_param.enable_body_fitting = True
b_param.detection_model = sl.BODY_TRACKING_MODEL.HUMAN_BODY_ACCURATE
b_param.body_format = sl.BODY_FORMAT.BODY_34


Could you record a SVO file (using ZED Explorer for example) that reproduces the issue and send it to ?

It would help us a lot.

In your configuration, the “z” axis is the forward direction, meaning the distance between the skeleton and the camera is incorrect?

Thank you for your response. I appreciate it. Here is the SVO file as requested. I have also sent a copy to the support email account.

In the SVO file, the user walks straight towards the camera in the Z direction, and the Z value increases with time. The issue that has made my customer and team aware of the problem is the drift/oscillation in the Z value. To reproduce this scenario, please view the 3D skeleton of the SVO file from a lateral view. The user should walk step by step, but it appears as if they are wearing ice skates.

that makes the consequence computation fail and incorrect.


The reply I got from the support team is this version of SDK is in the early stage-> means it is not stable.
here is the video of issue I encounter, low consistance in Z-direction of moving object.
for everyone’s reference.