Deepsort fail to assign same id to same people

Please provide complete information as applicable to your setup.

**• Hardware Platform (Jetson / GPU) RTX2060laptop
**• 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)question
• 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)

Currently i am using peoplenet as nvinfer(pgie), and the pipeline is based on deepstream-app.
The tracker model is original deepsort model and i am testing it with an rtsp ip camera.

One situation is when i am inside the images, the person id and face id can persist for a while, but when i covered my face for a second, the id of face changed and did not come back.
Another situation is when i walk out of the image and show up again, the face id and person id also changed.

So how to solve the problem? :0

For more detail, below is the config i used for deepstream-app to run.
you can easily reproduce the result if you have an ip camera.
source1_1080p_dec_infer-resnet_tracker_tiled_display.txt (4.4 KB)
config_tracker_DeepSORT.yml (5.0 KB)

1 Like

Is it possible to share one video file to reproduce the issue?


Sorry for late update.
Here is the video that can reproduce the issue.

Can you have a try with below setting for Deep SORT tracker?

maxShadowTrackingAge: 100
minMatchingScore4Overall: 0.4
minMatchingScore4SizeSimilarity: 0
minMatchingScore4Iou: 0
thresholdMahalanobis: -1

Sorry for late update.
We tried the parameters you provide and they did help improving the tracker’s performance.
Can you give us some guidence on how these parameters work so that we can do further fine tuning?
Thanks!

The target’s size and appearance changes a lot due to partial occlusion, so the minimal matching threshold should be lower. Also. increase maxShadowTrackingAge to store the target in shadow tracking mode longer.