This project implements video stabilization using OpenCV. It estimates motion between frames and smooths the trajectory using a Kalman filter, generating a stabilized video output with side-by-side comparison.

• Motion estimation via optical flow.
• Affine transformation model (translation + rotation).
• Kalman filtering to smooth the trajectory.
• Side-by-side visualization of original vs. stabilized video.
• Outputs transformation, trajectory, and smoothed data logs for analysis.

1. Feature Tracking: Good features are selected and tracked across frames using Lucas-Kanade optical flow.
2. Motion Estimation: Affine transform (translation + rotation) is computed between each frame.
3. Trajectory Building: The cumulative motion builds a raw trajectory.
4. Kalman Filtering: The trajectory is smoothed to reduce jitter.
5. Transform Correction: Differences between the raw and smoothed trajectory are used to compute new frame transforms.
6. Video Rendering: The corrected frames are shown side-by-side with original ones and saved as a new video (compare.avi).

• prev_to_cur_transformation.txt: Raw frame-to-frame transformations.
• trajectory.txt: Raw cumulative trajectory.
• smoothed_trajectory.txt: Kalman-filtered smooth trajectory.
• new_prev_to_cur_transformation.txt: Final stabilized frame transformations.

• Python 3.x
• OpenCV
• NumPy

pip install opencv-python numpy

python rec_stabi.py your_video_file.avi

You will see a window displaying original vs. stabilized frames side-by-side. The stabilized video is saved as compare.avi.

• HORIZONTAL_BORDER_CROP: Controls cropping to avoid black borders.
• Kalman filter parameters:
  • pstd: Process noise
  • cstd: Measurement noise

(Side-by-side comparison is shown in real time during execution.)

This project is open-source and available under the MIT License.

Developed using OpenCV and NumPy.