Wrong Id switch/assignment when using deepsort tracker in deepstream-6.1

Please provide complete information as applicable to your setup.

**• Hardware Platform (Jetson / GPU)GPU
**• DeepStream Version 6.1
• JetPack Version (valid for Jetson only)
• TensorRT Version
• NVIDIA GPU Driver Version (valid for GPU only)
• Issue Type( questions, new requirements, bugs)
• How to reproduce the issue ? (This is for bugs. Including which sample app is using, the configuration files content, the command line used and other details for reproducing)
• Requirement details( This is for new requirement. Including the module name-for which plugin or for which sample application, the function description)

Hello, i’m using deepstream-6.1 and deepsort tracker to sort person and face.
The model for nvinfer is peoplenet and the pipeline is developed in python.
During tracking person and face, i observed wrong id switch/assignment when using an ip camera.
There first show up a person and the tracker assigned an id for him. After the first person left ,the second person show up and the tracker seem to give the same id to the second person.
I reproduced this problem using an fake print picture and make a video.

I wonder this problem can be solved by adjusting some parameter in tracker yaml file, but i’m too shallow to figure it out, i really need some help.
Here is the tracker config yaml.
tracker_config.yml (3.2 KB)

Please have a try with latest DeepStream 6.2 release.
Please try to set checkClassMatch and useUniqueID for your issue.
Please share the test video without BBox and reproduce step, so we can check it in my side.

Thanks for you reply. I tested this video with deepstream-python-app/deepstream-test3 and it would give the same result of id switch. Just follow the readme and use peoplenet as nvinfer.
The provided tracker config yaml is the only thing i changed in the folder, i think you can easily reproduce my result.

The title of the topic say using deepsort tracker. But your tracker configure file is for NvDCF tracker. Which tracker are you using? Can you help to have a try with DeepStream 6.2 with this configure: deepstream/deepstream-6.2/samples/configs/deepstream-app/config_tracker_NvDCF_accuracy.yml? We improve tracker accuracy in DeepStream 6.2 release.

config_tracker_DeepSORT.yml (5.0 KB)

sorry for the wrong config file, you can use this file to reproduce the result.

hello, i tried deepstream-6.2 with nvdcf tracker in default configure file, but it seems unable to reidentify person when he reappeared, can you provide some guidance on how to set the parameter in the configure file?

Is it fix you issue in this topic?

For target reassociation, please refer: Gst-nvtracker — DeepStream 6.2 Release documentation

DeepStream trackers are implemented in such a way that when an object leaves the scene, the target tracker for that is immediately terminated and its states are removed. So, it is expected that when the person (either fake picture or real person) disappears from the scene and then re-appear their IDs will be re-assigned.

If the picture stays in the scene or the person stays in the scene and there are some brief occlusions, however, it will likely to be re-associated with the same ID.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.