NvDCF tracker update frequency

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

I’m using NvDCF for doing person detection, but sometimes I observe that if a person is walking at a fast pace, there’s a delay for the tracker bounding box to follow the person. As a result, sometimes it misses a small portion of the person body (I’m saving the person bounding box after detecting line crossing through nvdsanalytics) And when the person is walking slowly, the bounding boxes work perfectly fine. Is there a parameter in tracker_config.yml that controls this setting?

• Hardware Platform: T4
• DeepStream Version: 5.0
• TensorRT Version: 7
• NVIDIA GPU Driver Version (valid for GPU only): 440.33.01

3

Hello,

It seems related to the motion dynamics parameters. I would suggest to start with the following config values which allows fast-adapting states. These setting may result in less smoothed trajectories, but you can try lowering the values to find a good balance.

# [State Estimator] [MovingAvgEstimator]
trackExponentialSmoothingLr_loc: 0.9       # Learning rate for new location
trackExponentialSmoothingLr_scale: 0.9     # Learning rate for new scale
trackExponentialSmoothingLr_velocity: 0.9  # Learning rate for new velocity
4

Thanks for response I will try it out and let you know