This research prototype explores privacy-preserving preprocessing for autonomous driving datasets. It uses OpenCV and MediaPipe Face Mesh to triangulate faces, then warps triangle regions to reduce raw facial fidelity while keeping geometry consistent for downstream modeling.
The pipeline downloads a base face, extracts landmarks, applies the triangulation topology, and uses affine transforms to blend a new face region. Live webcam capture makes it easier to test the mesh transformation in real time and validate how much identity signal is removed.
- MediaPipe face mesh triangulation and affine warping.
- Live webcam capture with debug overlays.
- Focused on ethical AV dataset preprocessing.
- Python, OpenCV, MediaPipe, NumPy.
- Simple scripts for capture and mesh tooling.