Face Detection

gst-ai-face-detection performs face detection on each frame of a video stream, marking face locations and facial landmarks (eyes, nose, mouth, etc.).

Prerequisites

• Completed QIM SDK Installation and Model Download

Steps

1. Install ffmpeg and Transcode Video

The default video format has compatibility issues with Q900's GStreamer rendering pipeline. Transcode to baseline H.264:

radxa@airbox$

sudo apt install -y ffmpeg
sudo ffmpeg -y -i /etc/media/video.mp4 \
    -c:v libx264 \
    -profile:v baseline \
    -level 3.1 \
    -pix_fmt yuv420p \
    -vf scale=640:480 \
    -r 30 \
    -g 30 \
    -keyint_min 30 \
    -bf 0 \
    -an \
    -movflags +faststart \
    /etc/media/video_safe.mp4

2. Create Config File

radxa@airbox$

python3 -c "
import json
with open('/etc/configs/config_face_detection.json') as f:
    c = json.load(f)
c['file-path'] = '/etc/media/video_safe.mp4'
json.dump(c, open('/tmp/cfg_face_detection.json', 'w'), indent=2)
"

3. Run

radxa@airbox$

gst-ai-face-detection --config-file=/tmp/cfg_face_detection.json

Press Ctrl + C to stop.

Expected Output

Running app with model: /etc/models/face_det_lite_quantized.tflite and labels: /etc/labels/face_detection.json
VERBOSE: Replacing 90 out of 90 node(s) with delegate (TfLiteQnnDelegate) node
Pipeline state changed from PAUSED to PLAYING

The display shows the video with face bounding boxes and facial landmarks.

Validation

• 90 ops all delegated to DSP
• Pipeline reaches PLAYING state
• Display correctly shows face bounding boxes and landmarks