Face Recognition

Prerequisite

Nvidia Tx2 device with webcam.

Quick Start

1. Download docker image

docker pull myelintek/face_rec:v1.0-tx2

2. Check camera device

Check dvice path

ls /dev/video*

If target vdieo device are /dev/video1 . Modify scripts/start.sh and scripts/start-daemon.sh.

...
CAMERA_DEVICE=/dev/video1
CAMERA_DEVICE_NO=1
...

3. Start face recognition applicatinon

Run face recognition applicatin in foregrond mode.

bash scripts/start.sh

Run face recognition application in background mode.

bash scripts/start-daemon.sh

4. setup Grafana dashboard

Connect to http://localhost:3000 , login with default user admin, default password admin.

../../_images/grafana_1_login.png

Change password

../../_images/grafana_2_chpass.png

Go to “Config” > Data sources > Add Data sources

../../_images/grafana_3_config_ds.png ../../_images/grafana_4_add_ds.png

Add an influxdb data source

../../_images/grafana_5_add_ds.png

In HTTP scetion, specify URL http://localhost:8086

../../_images/grafana_6_ds_conf.png

In Influxdb detail section, specify database face_rec, then click “Save & test”

../../_images/grafana_7_ds_conf2.png

Go to “+” > Import > Upload JSON file

../../_images/grafana_8_import_1.png

Upload JSON file configs/face_rec_grafana_dashboard.json, then click import

../../_images/grafana_9_import2.png ../../_images/grafana_10_import_3.png

5. Start Monitoring

../../_images/grafana_11_dashboard.png

  • Face log : When and who detected

  • Face count: Count in time period.

  • Performance: face recognition performance.

Use you own model

1. Prepare training dataset

Folder structure, use class name as folder name. each folder contain one people’s picture.

images/
    name_01/
        pic1.jpg
        pic2.jpg
    name_02/
        pic1.jpg
        pic2.jpg
    ...

2. Train face classification model

bash scripts/train.sh <image_dir> <output_dir>

  • image_dir: training image directory.

  • output_dir: director for save trained model file

3. Replace model

Trained model should contain encoderSVM.pkl and classifierSVM.pkl. Then mount model_dir to /face-recognition/models folder in container to override the deafult model. Add following parameter when you start container.

bash scripts/start.sh your/model_dir
bash scripts/start-daemon.sh your/model_dir