Real-time video stabilization

Example

[NOTE] Now works ONLY stabilizer package. stabilizer_v2 is in development!

About

This project provides two Python packages for real-time digital image stabilisation for the OAK-D Pro camera.

Stabilizer

These packages use feature detection techniques for key point tracking. Based on key points, create 3x3 homography rotation matrix. Then using running average filter, smooth frame path and warp input frame.

Stabiizer_v2

This package will work on quaternion rotation vector using data from MEMS-sensor in camera. This package is currently under development.

How run

1. Install all library's and packages

[NOTE] User's PC must have already installed python 3.11 or higher.

pip install -r requirements.txt

2. Connect camera to PC.
3. Run application.

[NOTE] If you don't want to run stabilisation in preview mode (show original and stabilised video, but in cropped resolution), change the preview_mode parameter in main.py to False.

python main.py

4. Close window on q or esc