Skip to content

Latest commit

 

History

History
112 lines (87 loc) · 3.71 KB

README.md

File metadata and controls

112 lines (87 loc) · 3.71 KB
Raw
       _      _             _____ ____  
      (_)    (_)           |_   _/ __ \ 
 _   _ _  ___ _  ___  _____  | || |  | |
| | | | |/___) |/ _ \|  _  \ | || |  | |
 \ V /| |___ | ( (_) ) | | |_| || |__| |
  \_/ |_(___/|_|\___/|_| |_)_____\____/

VisionIO

The VisionIO module enables ravestate to connect to a Face Recognition ROS topic provided by face_oracle, based on which conversations can be initiated.

Architecture

OpenCV Video Stream
----> (     FaceOracle Client   )
      (webcam_video_processor.py) ---> ravestate_visionio
        |                     A        /roboy/cognition/vision/visible_face_names
        |                     |        (ROS1)
        V                     |
  face_recognition            |
        |              name-confidence pairs
        V                     |
  facial feature vector       |
        |                     |
        V                     |
     (FaceOracle Websocket Server)
        |                     A
        V                     |
  Load FaceVector-PrimKey     |
  pairs from Redis            |
        |                     |
        V                     |
  Match with Request          |
  Vector via Pyroboy          |
  FaceRec.match_face          |
        |                     |
        V                     |
   [ Name for PrimKey from Scientio ]

Dependencies

VisionIO requires the following components to be running:

  • ravestate or raveboard with ravestate_visionio module
  • Neo4j backend for Scientio
  • redis for persisting facial feature vectors
  • face_oracle client and server

Configuration

VisionIO provides the following config keys:

Key Default Description
redis_host Host for Redis database. localhost
redis_port Port for Redis database. 6379
redis_pass Password for Redis database Empty
ros1-node Topic for Faces messages. /roboy/cognition/vision/visible_face_names
min-confidence Minimum confidence below which someone will be a stranger. 0.85

How to run

We recommend running VisionIO through one of the ravestate docker-compose profiles, which will start Neo4j, redis, and the face_oracle client and server automatically.

Start the profile and visionio in docker as follows:

> docker-compose up -d {profile}
> docker exec -it rs bash
> python3 -m ravestate ...

To start face recognition, open localhost:8088/index.html in a browser (Chrome works best). This will give a visualisation of recognised faces, and simultaneously keep recognition running. Face recognition will only work as long as you can see it doing so!

The docker-compose profiles differ per operating system:

Profile rs-linux

On Linux, a Webcam for VisionIO can simply be mapped into docker as a device. Per default, this will be video0.

If you want to change the device, map a new device in docker-compose.yml, and don't forget to change the FACEORACLE_VIDEO_DEVICE variable.

Profile rs-macos

On Mac, Docker can not natively access USB devices. Instead, live video can be streamed into the container via RTMP:

  1. Install/start Local RTMP Server
  2. Install ffmpeg via brew install ffmpeg
  3. Stream webcam via RTMP by starting ravestate/run_ffmpeg_stream.sh

You can now start docker-compose up -d rs-macos.

Using a video instead of a webcam feed

If you don't have a webcam, you can use a video instead for debugging. Just set FACEORACLE_VIDEO_DEVICE for your particular platform profile to /ravestate/resources/obama.mp4. Note, that after changing docker-compose.yml, you have to run docker-compose up -d with the --force-recreate flag.