Human Pose Estimation and Dynamic Parameter Adjustment with ZED 2 Camera

Software C++
1

I am currently working on a project involving human pose estimation using the ZED 2 camera. I have encountered a couple of issues that I hope you could assist me with:

  1. Tracking ID Shifting During Occlusion: While tracking a specific person, I have observed that if two or more people are occluded, the track ID tends to shift after 1 or 2 seconds sometimes and sometimes not. This leads to inconsistencies in tracking the intended person. Could you please provide guidance or best practices to ensure consistent tracking of a specific individual, even during occlusions? instead of shift it needs to change or incremented.
  2. Dynamic Adjustment of Detection Confidence Threshold: I am also looking to dynamically adjust the body_tracker_parameters_rt.detection_confidence_threshold parameter during the runtime of the application. Currently, the threshold is set statically before the loop begins, and I need to modify it dynamically based on certain conditions within the application loop. Could you provide an example or documentation on how to achieve this?Thank you for your time and assistance.
2

Hi @debanik123,

  1. The occlusion management is a known limitation of our tracking, you can tinker with the body tracking runtime parameters to attenuate the switching. Mainly, look into the minimum_keypoints_threshold and
    detection_confidence_threshold    .
  2. You can change the runtime parameters at runtime, so just set your conditions inside the loop and change the value of the body_tracker_parameters_rt’s variables.
3

hii,
Thank you :blush:.

Thanks and Regards
Debanik Roy