balena-io-experimental / sense-people

Count people passing by, display it nicely with a Grafana dashboard.

People counter + balenaSense demo

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.

Required hardware

• 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

Before you start

1. Plug the USB webcam (unless you are using an IP camera) and the USB WiFi dongle into the Jetson Nano
2. Fit a jumper to J48 on the Jetson Nano board (instructions)
3. Flash the SD card, and you are good to go!

Usage: Visualization

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.

Usage: Change object detection model

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

Usage: Use an IP camera

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:[email protected]

sense-people's Issues

update project to use newest edgeiq:1.8.0 release?

https://alwaysai.co/docs/edgeiq_api/edgeiq_release_notes.html#release-1-8-0

My attempt to update the edgeiq version from alwaysai/edgeiq:nano-0.11.0 to alwaysai/edgeiq:jetson-1.8.0 was not successful

Any plans to update this project to use jetpack4.6 and alwaysai/edgeiq:jetson-1.8.0 ?

Here is my modified Dockerfile

#FROM alwaysai/edgeiq:nano-0.11.0
FROM alwaysai/edgeiq:jetson-1.8.0
WORKDIR /usr/src

COPY . .
RUN apt update && apt install -y python3-pip
RUN pip3 install paho-mqtt

CMD [ "python3", "app.py" ]

My attempt resulted in the following error. Does this have anything to do with not incorporating the alwaysai venv which seems to want python3.7 instead of python3.6?

people-counter-1.8_1  | Traceback (most recent call last):
people-counter-1.8_1  |   File "app.py", line 10, in <module>
people-counter-1.8_1  |     import edgeiq
people-counter-1.8_1  |   File "</usr/local/lib/python3.7/site-packages/edgeiq/__init__.py>", line 3, in <module>
people-counter-1.8_1  |   File "<frozen edgeiq>", line 12, in <module>
people-counter-1.8_1  | Exception: Invalid token! Please run "aai app configure" or "aai user login" to refresh the token. (Error code: 219)

eclipse-mosquitto:1.6.15 mqtt image seems to work

Thanks for this wonderful project and associated blog post https://www.balena.io/blog/build-an-ai-driven-object-detection-algorithm-with-balenaos-and-alwaysai/
I had to make the following changes to the docker-compose file to get this running on a jetson-nano
arm32v6/eclipse-mosquitto:latest seems to not work. When I switched to eclipse-mosquitto:1.6.15, that did the trick
balena-labs-projects/balena-sense@48b75a8

mike@nano159:~/IOTstack/sense-people$ git diff docker-compose.yml
diff --git a/docker-compose.yml b/docker-compose.yml
index f5aeec9..2425988 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -11,11 +11,11 @@ services:
     restart: always
     build: ./balena-sense/grafana
     ports:
-      - "80"
+            - "3000:3000"
     volumes:
       - 'sense-data:/data'
     environment:
-        - 'GF_SERVER_HTTP_PORT=80'
+        - 'GF_SERVER_HTTP_PORT=3000'
         - 'GF_PATHS_PROVISIONING=/usr/src/app/provisioning'
         - 'GF_SESSION_PROVIDER=memory'
   telegraf:
@@ -24,7 +24,7 @@ services:
     cap_add:
       - SYS_ADMIN
   mqtt:
-    image: arm32v6/eclipse-mosquitto
+    image: eclipse-mosquitto:1.6.15
     ports:
       - "1883:1883"
     restart: always

grafana: IP address for alwaysAI stream is hardcoded

See: https://github.com/balena-io-playground/sense-people/blob/master/balena-sense/grafana/provisioning/dashboards/balena-sense.json#L140

Need to change that