Tutorial 13 — Face Blur Privacy Pipeline with Jetson

Automatically detect and blur all faces in a video stream in real time — useful for GDPR compliance, public-space cameras, research datasets, and any application where identity must be protected without losing the utility of the video.

What you will learn

• How to detect faces at 60+ FPS using a lightweight YOLOv8 face model
• How to apply Gaussian blur, pixelation, or black-box anonymisation
• How to process and save anonymised video files offline
• How to balance detection accuracy vs processing speed
• GDPR and privacy considerations for AI vision systems

Step 1 — Run the face blur demo

cd ~/tutorials/13-face-blur
python3 face_blur.py --source 0 --method gaussian --show

Step 2 — Three blur methods

def blur_region(frame, x1, y1, x2, y2, method='gaussian'):
    roi = frame[y1:y2, x1:x2]

    if method == 'gaussian':
        blurred = cv2.GaussianBlur(roi, (51, 51), 30)

    elif method == 'pixelate':
        h, w   = roi.shape[:2]
        temp   = cv2.resize(roi, (w//10, h//10))
        blurred= cv2.resize(temp, (w, h), interpolation=cv2.INTER_NEAREST)

    elif method == 'blackout':
        blurred = np.zeros_like(roi)

    frame[y1:y2, x1:x2] = blurred
    return frame

Step 3 — Process and save a full video file

python3 face_blur.py     --source ~/videos/original.mp4     --method pixelate     --output ~/videos/anonymised.mp4

✅ Next: Tutorial 14 — YOLO11 Benchmark | Back to Jetson Kit