Using nvTracker lib's in pure TensorRT app

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Hi all!

I would like to use the low-level tracker .so’s provided in Deepstream (eg: libnvds_mot_klt.so) in my non-Gstreamer app which is built using only TensorRT. Could you please guide me on how I can use this for object tracking and unique ID assignment? I will be performing detections with YoloV3 which is already implemented.

Any help would be greatly appreciated!

2

Hi,

Our tracker is GStreamer-based implementation so you will need to prepare GStreamer-type input/output.
A sample for launching a gst component can be found here:
/opt/nvidia/deepstream/deepstream-4.0/sources/apps/sample_apps

For your use case, you can check this topic first:
https://devtalk.nvidia.com/default/topic/1066252/klt-nvmot-usage/

The user want to leverage NvTracker but get stuck in the memory type.
Although we are still checking the possible cause, it should be able to give you some information.

Thanks.