Status: Conceptual work
(beware really 1st shot, only wanted to document it)
Base: AI on the edge => AI on the server
Task: Generate information out of image segements using OCR (tensorflow-lite) and tesseract
Workflow:
- Generate Configuration (Areas on the image, which conntain information)
- Send Image to OCR Server
- Image areas are interpreted via tensorflow or tesseract
- Extracted Information is returned via JSON or forwarded to Mqtt
First running HAOS app
based on OCR-Server build for docker
1st Test:
curl -s -X POST <HAOS-IP>:5000/segment -F "identifier=watermeter" -F "[email protected]"
{"identifier":"watermeter","results":[
{"id":"d1","value":"00016.0"},
{"id":"a1","value":9.127448058121708},
{"id":"a2","value":3.881748198574262},
{"id":"a3","value":1.2311287412814798},
{"id":"a4","value":3.345734090629815},
{"id":"t1","value":"CH-MI001-10039"}]}
Configuration Editor for image segments:
http://HAOS-IP:5000
Changes:
- 0.1.2 : Initial take over as HAOS add-on/app
- 0.1.4 : MQTT parameter available/overwriteable under options