Petrify
ActiveFreeze the world. Keep the clock ticking.
An open source computer vision tool that removes people from camera footage while preserving the timestamp overlay. When someone enters the scene, the background freezes into a clean still and only the clock keeps moving. Frame-perfect, zero-config, and it runs fully on your own machine.
What you can do with it
Privacy compliance
Strip identifiable individuals for PDPA / GDPR while keeping scene context and a verified timeline.
Footage compression
Drop sparse motion (a bird, a leaf, a flag) and keep only the static background. Much smaller files.
Space & layout analysis
Remove people occluding a floor so you can build clean heatmaps and CAD-ready layouts.
Hazard detection
Expose obstacles, spills and trip points by clearing human presence from the frame.
Clean AI training data
Strip subject identity from annotated scenes while preserving the surrounding environment.
Any fixed camera
General foreground removal for a clean background plate with a live, honest timestamp.
How it works
Petrify watches the frame for foreground subjects and swaps the live background for a clean, pre-selected still the moment one appears. The timestamp region is detected automatically and keeps updating in real time, so the footage stays verifiably current without ever showing a person. The scene says nothing happened here. The timestamp says time kept moving.
- Frame-perfect person detection and removal.
- Automatic timestamp region detection, no configuration needed.
- Fully local processing with no cloud dependencies.
- Three-detector pipeline: MOG2 motion, HOG person detection, static background differ.
- Hardware acceleration on Apple Silicon and NVIDIA GPUs.
Stack & status
Python 3.10+ with OpenCV, FFmpeg, NumPy and tqdm. Processes 1080p at roughly real time on Apple M-series chips, faster at lower resolutions. Apache 2.0 licensed, in active development.
Use it responsibly
Always make sure you have the legal right to process any footage you use. Petrify exists to balance security and privacy, not to undercut either.