Using separate threads for displaying ip camera stream and performing tracking on Jetson Xavier NX

[canozcivelek]

Hi, I’m performing visual object tracking on a stream from an IP camera. I constructed a Gstreamer pipeline with Nvidia acceleration and it’s working great (low latency, high fps). However, when I added a few tracking algorithms on top of the stream, things got jerky and I’m looking to improve performance of tracking. I’m on a Jetson Xavier NX, Python 3.6, and would like to know what I can do to improve performance while tracking. I’m thinking of separating the tracking from streaming and use threading or multiprocessing but couldn’t manage to find a solid application. Any suggestions? Thanks a lot!

[DaneLLL]

Hi,
The optimal solution of runnig deep learning inference is DeepStream SDK.

Announcing DeepStream 5.0.1

Announcing availability of DeepStream 5.0.1 on Devzone and NGC. DeepStream 5.0.1 fixed few bugs that our users found in DeepStream 5.0. Here’s a list of the bugs that we addressed in DeepStream 5.0.1 Demuxer plugin – Stream content were mixing after nvstreamdemux Dewarper plugin – Fixed aspect ratio Python binding – Fixed bindings for NvDsLabelInfo The Demuxer plugin bug has been reported by several users; please upgrade to DeepStream 5.0.1 to pick up the latest fixes. Here are few issues re…

Would suggest you check the sample apps and give it a try.

[canozcivelek]

Thanks for the quick reply! I’ll give it a try. However, I’d still appreciate some ideas on implementing some sort of a threaded processing of frames and performing tracking on parallel. Thanks!

[DaneLLL]

Hi,
The implementation is to run a pipeline sequentially. If you would like to run the models parallelly, may need to customize the default implementation. For object tracking, we have nvtracker plugin. You may give it a try and see if it can be applied to your usecase.