Accessing CCTV camera feeds

We are creating computer vision systems and would like to process multiple video feeds at high resolution from CCTV cameras.

We would like to know the best practices to access CCTV video feeds via ethernet on the Jetson Xavier NX platform.

  1. Is RTSP stream the best way to get real time frames from the CCTV camera?
  2. What are the best methods to access the camera in Python?
  3. Currently, when using OpenCV’s video capture and we notice a long delay (~7 seconds) in receiving frames from main feeds of CCTV cameras. Can this be avoided?
  4. Are are any samples to implement Deepstream for capturing CCTV video feeds?

Please let us know. Thanks.

Not sure which type CCTV is. Generally we use RTSP source and V4l2 source on Jetson platforms. For python samples, please refer to
Doesn't work nvv4l2decoder for decoding RTSP in gstreamer + opencv - #3 by DaneLLL
V4l2src using OpenCV Gstreamer is not working in Jetson Xavier NX - #3 by DaneLLL

You may try the samples to capture frame data from CCTV. And if it works, it shall be working the same in DeepStream SDK. The pythson samples for DeepStream SDK are in
GitHub - NVIDIA-AI-IOT/deepstream_python_apps: DeepStream SDK Python bindings and sample applications


Thank you for your response. We are able to run the Deepstream sample apps. However, we are looking for face detection in the frames of multiple CCTV RTSP streams. Are there any specific models we can use for this or any sample application that would run on Jetson Xavier NX? Please let us know.

There are some models in DeepStream SDK for demonstration. Would see if other users can suggest a model for face detection.


  1. Every sample we see does not have a model built for the Jetson Xavier NX. Is there a model zoo or similar repository of models that are prebuilt for the Xavier NX? The process of building or modifying the model for this platform seems to take a long time and not very straight forward.

  2. Is there a better way to try different samples with Deepstream without breaking the environment running other programs? Like using dockers, etc.? Please suggest.



You can find some pre-trained models on our NGC website.
The TAO-based model should be workable on Jetson.

Building takes time since it optimizes based on the hardware info.
But it is a one-time job.

Deepstream docker is also available on the NGC website.


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