Open DRUWA

Open Deep Realtime User Welcoming Assistant

Realtime face detection & verification using Webcam.

contact: [email protected]

AIRI - AI Research Institute

Requirement

- cmake / gcc,g++ / python 2.7 / pip
- CUDA/CUDNN
- REDIS Server / hiredis / PyRedis 
- OpenCV2 or OpenCV3 for support CUDA
	- Note) You can install libopencv-dev. But cpu only.
- Darknet
	- C/C++ dnn framework
- Openface
	- You can user docker image which does not support gpgpu
- Torch
	- Case of building Native Openface 
- dlib

CUDA/CUDNN

- NVidia Wed Site.
- CUDA 8.0 / CUDNN 5.1(tested), CUDNN 6.0(tested) 

Installation

    Change current directory to RealtimeCamera
    Run install_packages.sh
    Run setup.sh

Run

    Run Redis Server
        Admit external access(Edit redis.conf)
        Remember server ip and port
        
    Run face_detector/run_camera.sh
        Edit run_camera.sh argument
        bin/darknet yolo camera cfg/yolo-face.cfg models/face.weights this_server_ip redis_server_ip


    Edit face_recognizer/cascaded_recognizer.py
        HOST, PORT = "10.100.1.152", 55555 -> face_detector server
        REDIS_SERVER = '10.100.1.150'      -> Redis Server
        REDIS_PORT = 6379
        
    Run face_recognizer/cuda_recognizer.sh

Directory Structure

    face_detector
        extractor
            YOLO based on darknet
            Camera Control based on OpenCV
            Face data redis publish
        identifier
            Person label(eg.name) selection
            Face tracking
    face_recognizer
         Face data redis subscriber
         Face alignment with dlib python
         Vector distance checker
         DBN classifier inference
         SVM classifier inference
         TTS engine(Google Voice)
    face_register
         Supervised image training
         DBN with LDA dimension reduction
         SVM(RBF) fine tunning
    robot
         Arduino controller(Door open)
         node.js event receiver

Training

    Input data
        Place face images in face_register/input/user
        input/user
            First target class
        input/iguest
            Second target class
            
    Run face_register/train.sh
    
    TDB

Enjoy.