Work with pi camer?

[cozmo14047]

Hi there does this work with araspberry pi also this uses obicos api to ai detect ont he camera stream right.
Manu thanks

[nvtkaszpir]

can you rephrase your question please?
or explain what you have and what you want to do, and which subdirectory are you referring to?

[cozmo14047]

Hi there sorry, i didint realise it was in just 3d printing direcory, for the obico one i see you say esp but nothing about raspberry pi. i just wanted to know if its compatable with a raspberry pi many thanks

[nvtkaszpir]

You would have to expose rpi camera snapshot as static image over the web (there are many ways how to do it), that would be the best option.

Using camera device directly would require more code adjustments (using python picamera), but because it would require using camera directly via python it may interfere with other tools - so I'm not going to do it.

[nvtkaszpir]

I think the easiest would be to run some process which would expose rpi camera as snapshot and then use that endpoint to fetch image.
Short search gives that https://github.com/signag/raspi-cam-srv should work with https://<server>:<port>/photo_feed

[nvtkaszpir]

if you mean you want to use ml_api on the rpi and do detections on the image fetched directly from the camera, then no, right now this is not possible and it would require much more work:

• add code to fetch image from camera using picamera python lib
• compile code for rpi (arm and arm64 platfroms) which means fighting a lot fo dependencies
• compute on rpi would be pretty slow and probably would take seconds to do a detection on full screen image.

For now this is just too much work so I'm not planning to do it.
Closing.

[nvtkaszpir]

try https://github.com/nvtkaszpir/3d-print/tree/main/picamera-web as an endpoint exposing pi camera image over the web

[cozmo14047]

Thank you. I didint think it would work with the pi thats why i asked

[nvtkaszpir]

what I tested:

Running picamera-web on rpi and obico and scripts on another host

• run https://github.com/nvtkaszpir/3d-print/tree/main/picamera-web on rpi 4, it exposes rpi camera on 192.168.1.50:8090

• on another computer (preferably with amd64) run obico ml-api only https://github.com/nvtkaszpir/3d-print/tree/main/obico-ml-api-only
  and run

  cd   obico-ml-api-only
  docker-compose up
  python3 draw_detections.py --api https://127.0.0.1:3333 https://192.168.1.50:8090/ --show

• as a result obico fetched image from 192.168.1.50:8090 and ran detection on it

Testing it directly on rpi4 on 64bit Debian 12, where 192.168.1.50 is the IP address of the raspberry pi:

• run all above on the rpi 4 - this way it would run obico and camera exposing service on the same device

• it should work but will be slow

• install docker and obico ml-api container (notice it fetches 3GB so it will take time)

   sudo docker run -it -p 3333:3333 quay.io/kaszpir/ml_api:812a05b7-arm64 gunicorn --bind 0.0.0.0:3333 --workers 1 wsgi --preload -e DARKNET_PATH=/darknet

• in another terminal, fetch scripts and run picamera-web

   cd src
   git clone https://github.com/nvtkaszpir/3d-print.git
   cd 3d-print/picamera-web
   make web

• in another terminal

   cd src/3d-print/obico-ml-api-only
   python3 draw_detections.py --api https://127.0.0.1:3333 https://192.168.1.50:8090/ --show

• result:

   pi@hormex:~/src/3d-print/obico-ml-api-only $ Error: no "view" rule for type "image/png" passed its test case
          (for more information, add "--debug=1" on the command line)
   python3 draw_detections.py --api https://127.0.0.1:3333 https://192.168.1.50:8090/
   INFO:root:treshold=0.2
   INFO:root:api=https://127.0.0.1:3333
   INFO:root:ignore=[]
   INFO:root:show=False
   INFO:root:img_url=https://192.168.1.50:8090/
   INFO:root:saveimg=None
   INFO:root:savedet=None
   INFO:root:ignore_list: []
   DEBUG:urllib3.connectionpool:Starting new HTTP connection (1): 192.168.1.50:8090
   DEBUG:urllib3.connectionpool:https://192.168.1.50:8090 "GET / HTTP/1.1" 200 54196
   DEBUG:urllib3.connectionpool:Starting new HTTP connection (1): 127.0.0.1:3333
   DEBUG:urllib3.connectionpool:https://127.0.0.1:3333 "GET /p/?img=http%3A%2F%2F192.168.1.50%3A8090%2F&ignore=%5B%5D HTTP/1.1" 200 74
   INFO:root:detections: [["failure", 0.119, [326.0, 133.0, 264.0, 326.0]]]
   INFO:root:ignored: []
   
  

it detected failed area with confidence 0.119 (11.9%) - mainly because I have there some dummy spaghetti in front of camera :)

So it works but is slow.

[nvtkaszpir]

also, obico-ml-api-only was not tested with video streams (I doudb it would work), so you would have to make a new service to expose single camera image from a stream so that obico ml api could fetch it and process it.
For such scenarios I suggest just using official obico.