This project combines balenaSense with the people counter demo from alwaysAI. It creates a grafana dashboard that shows the video stream from alwaysAI with aditional metrics.
- Jetson Nano Developer Kit
- A DC barrel jack power supply with 5.5mm OD / 2.1mm ID / 9.5mm length, center pin positive that can supply up to 5V⎓4A
- A jumper to short J48
- A USB Webcam or an IP camera which supplies an RTSP feed
- A WiFi USB dongle
- Plug the USB webcam (unless you are using an IP camera) and the USB WiFi dongle into the Jetson Nano
- Fit a jumper to J48 on the Jetson Nano board (instructions)
- Flash the SD card, and you are good to go!
To visualize statistics and a video stream:
- Grafana: http://deviceLocalIP
- alwaysAI: http://deviceLocalIP:5000
Note that the grafana dashboard will show both stats and the alwaysAI stream so it's better for a live demo.
You can change the model that is used to perform object detection via the OBJECT_DETECTION_MODEL environment variable:
| Neural Network Framework | Dataset | Model | OBJECT_DETECTION_MODEL value | Reference inference time |
|---|---|---|---|---|
| caffe | COCO | MobileNet SSD | mobilenet_ssd (default) | 400 msec |
| darknet | COCO | Yolo v2 tiny | yolo_v2_tiny | 60 msec |
| darknet | VOC0712 | Yolo v2 tiny | yolo_v2_tiny_voc | 70 msec |
| darknet | COCO | Yolo v3 tiny | yolo_v3_tiny | 60 msec |
You can add an RSTP feed url via the IP_CAMERA_FEED environment variable. Examples:
rtsp://192.168.1.10:88
With a video stream defined:
rtsp://192.168.1.10:88/mainVideo
Feed with authentication:
rtsp://username:password@192.168.1.10