OpenVino Face Recognition modle is wrapper of openvino toolkit with Intel Movidius for Azure Video Analyzer on Edge on Raspberry Pi.
In a pipeline of AVA on Edge, you can use following features.
| Feature | URI |
|---|---|
| Face Recognition | http://ipaddress:8888/facerecognition |
| Object Detection | http://ipaddress:8888/objectdetection |
This sample is based on information published at the following sites.
| HW | Raspberry Pi 4 8GB Ram |
| OS | Raspbian Arm32v7 |
※ You can use Raspberry Pi 3 but you should spend long long time for build. In the case of Raspberry Pi 4 8MB Ram, it takes about 3 hours.
By SSH shell, execute following command
git clone https://github.com/ms-iotkithol-jp/openvino-face-recognition-with-ava-on-edge.git
cd openvino-face-recognition-with-ava-on-edge/EdgeSolution/modules/openvinofacerecog
sudo docker build -t openvinofacerecog -f Dockerfile.arm32v7
sudo docker tag openvinofacerecog your-docker-hub/openvinofacerecog:version-arm32v7
sudo docker push your-docker-hub/openvinofacerecog:version-arm32v7You can use Raspberry Pi 3 Model B or Raspberry Pi 4. Installed OS should be Raspbian ARM64.
Please setup AVA on Edge and Blob on Edge according to following Docs site.
Make directory for Face Identity Database.
sudo mkdir /facedbPlease add several jpeg files of persons who you want to idenfiy according to https://github.com/openvinotoolkit/open_model_zoo/tree/master/demos/face_recognition_demo/python
You can use various AI learned model for this feature.
Available models are listed at https://github.com/openvinotoolkit/open_model_zoo/tree/master/demos/object_detection_demo/python#supported-models
For example you'd like to use yolo-v2-ava-0001, download the model files and format them into a gzipped tar file on your Raspberry Pi shell.
cd ~
mkdir odmodel
git clone https://github.com/openvinotoolkit/open_model_zoo.git
cd open_model_zoo/tools/downloader
python3 -mpip install --user -r ./requirements.in
python3 downloader.py --name yolo-v2-ava-001
cp intel/yolo-v2-ava-0001/FP16/* ~/opmodel
cd ~/opmodel
cp open_model_zoo/data/dateset_classes/voc_20cl.txt
cd opmodel
tar zcvf od-model.taz yolo-v2-ava-001.xml yolo-v2-ava-001.bin voc_20cl.txtOpenVino Face Recognition logic use following models
| Usage | Model |
|---|---|
| Face Detection | face-detection-retail-0004 |
| Landmark Regression | landmarks-regression-retail-0009 |
| Face Identification | face-reidentification-retail-0095 |
| Age Gender Detection | age-gender-recognition-retail-0013 |
Download above models and format them into gzipped tar file.
cd ~/open_model_zoo/tools/downloader
python3 downloader.py --name face-detection-retail-0004
python3 downloader.py --name landmarks-regression-retail-0009
python3 downloader.py --name face-reidentification-retail-0095
python3 downloader.py --name age-gender-recognition-retail-0013
mkdir ~/frmodel
cp intel/face-detection-retail-0004/FP16/* ~/frmodel
cp intel/landmarks-regression-retail-0009/FP16/* ~/frmodel
cp intel/face-reidentification-retail-0095/FP16/* ~/frmodel
cp intel/age-gender-recognition-retail-0013/FP16/* ~/frmodel
cd ~/frmodel
tar zcvf fr-model.tgz face-detection-retail-0004.* landmarks-regression-retail-0009.* face-reidentification-retail-0095.* age-gender-recognition-retail-0013.*Next create face database for reidentification model.
- Prepare images taken from the front of the face for each person in JPEG format and store them into same folder.
- Create gzipped tar file including the face image files.
tar zcvf facedb.tgz john.jpg bob.jpg ...After created tgz file, upload it to Azure Blob Container and create URL with SAS Token for downloading by IoT Edge Module.
{
"NetworkingConfig": {
"EndpointsConfig": {
"host": {}
}
},
"HostConfig": {
"Binds": [
"/dev:/dev",
"/facedb:/facedb"
],
"Privileged": true,
"PortBindings": {
"8888/tcp": [
{
"HostPort": "8888"
}
]
},
"NetworkMode": "host",
"ExposedPorts": {
"8888/tcp": {}
}
}
}{
"send-telemetry": true,
"upload": {
"interval-sec": 70,
"inference-mark": true
},
"model-od": {
"url": "<- url of object-detection-model.tgz ->",
"filename": "od-model.tgz",
"name": "yolo-v2-ava-0001.xml",
"label": "voc_20cl.txt",
"architecture-type": "yolo"
},
"model-fr": {
"url": "<- url of face-recognition-related-models.tgz ->",
"filename": "fr-model.tgz",
"fd-name": "face-detection-retail-0004.xml",
"lm-name": "landmarks-regression-retail-0009.xml",
"reid-name": "face-reidentification-retail-0095.xml",
"ag-model" : "age-gender-recognition-retail-0013.xml"
},
"face-db": {
"url": "<- url of face-database.tgz ->",
"filename": "facedb.tgz"
}
}For the case of Raspberry Pi + INTEL Movidius
| Name | Value |
|---|---|
| BLOB_ON_EDGE_MODULE | name of the deployed 'Blob on Edge' module |
| BLOB_ON_EDGE_ACCOUNT_NAME | local account name specifyed in 'Blob on Edge' module |
| BLOB_ON_EDGE_ACCOUNT_KEY | local account key specifyed in 'Blob on Edge' module |
| BLOB_CONTAINER_NAME | container name specifyed in 'Blob on Edge' module |
| FACEDB_FOLDER_NAME | folder name of face database of face recognition. the folder should be same of create option specified |
| OPENVINO_DEVICE | MYRIAD |
| MSG_OUTPUT_PATH_OD | message output path for object detection. ex) output_od |
| MSG_OUTPUT_PATH_FR | message output path for face recognition. ex) output_fr |
When you want to use this IoT Edge Module quickly. You can use embeddedgeorge/openvinofacerecog:0.5.1-arm32v7 .