The IntelligentEdgeHOL walks through the process of deploying an IoT Edge module to an Nvidia Jetson Nano device to allow for detection of objects in YouTube videos, RTSP streams, or an attached web cam.
It is available as a Github Repo that can be cloned @ https://github.com/toolboc/IntelligentEdgeHOL
The project uses Azure IoT Edge https://docs.microsoft.com/en-us/azure/iot-edge/about-iot-edge?WT.mc_id=nvidia-IntelligentCCTV-pdecarlo to deploy a containerized module which runs Tiny Yolo v3 using a specially tuned fork of Darknet that has been compiled to take advantage of onboard CUDA libraries for GPU acceleration. GPU acceleration is achieved within the container using the strategy detailed in this blog post https://dev.to/azure/supercharge-your-containerized-iot-workloads-with-gpu-acceleration-on-nvidia-jetson-devices-4532. This allows us to achieve object detection of at around 11 FPS using the Jetson Nano with an off the shelf Logitech Webcam.
The project makes it very easy to configure for use with a variety of video sources, making this a great project for demonstrating the capabilities of the Nvidia Jetson Nano device. You can easily add a link to any publicly available Youtube URL and begin performing object detection on that video. The inferencing output is served over an http server which allows you to view the object detection results live from a connected web-browser so that you can view results without the need for a physically attached display.
Take a look at the repository linked at the beginning of this post to get started.