Introduction

This is an implementation of a video stabilization system using ORB descriptor. The input of the program is a video, and it outputs the stabilized video in the same folder.

Requirements

The program was written in Python. The libraries used are:

• OpenCV 3.1 (Use these instructions: http://docs.opencv.org/2.4/doc/tutorials/introduction/linux_install/linux_install.html#linux-installation)
• Numpy 1.8.2
• SciPy 0.13.3
• Matplotlib 1.3.1

To run

to run the program simply use:

python videoStab.py [video]

Parameters must be changed manually in videoStap.py

• videoInPath: path of the unstabilized video (optional)
• MATCH_THRES: Matches distance threshold
• RANSAC_THRES: RANSAC threshold
• BORDER_CUT: Number of pixel to crop in output video
• FILT: filter type (square or Gauss)
• FILT_WIDTH: filter width
• FILT_SIGMA: filter variance (only in Gaussian filter)
• FAST: If true use the fast version of the algorithm

Refecences

[1] Xu, Jie, et al. "Fast feature-based video stabilization without accumulative global motion estimation." IEEE Transactions on Consumer Electronics 58.3 (2012). [https://ieeexplore.ieee.org/document/6311347/?arnumber=6311347]