Motion Compensation using Affine warp

Introduction

The aim of this project is to compensate for the motion in a video. Intutively, we first detect the motion and then apply a transformation that is opposite to the detected motion. The fooling are the steps followed:

• Detect feature points
• Track those feature points
• Estimate motion
• Apply inverse transformation

I used Lucas-kanade tracter to track the feature points.

Building and running the code

• Clone the github repository
  git clone https://github.com/sakshikakde/video_motion_compensation.git

• Go to the repositiory folder cd video_motion_compensation and create a build directory mkdir build. Go to the build directory cd build.

• Run the following commands to build the project.
  cmake ..
make

• The executable will be generated in the build folder. To run the program, use the following command.
  ./VideoStabilization ../dataset/motion_video2.mp4

Results

Original input video

Motion compensated video

Improvement scope

I am using an affine warp for this example. As it can be seen, the output is a little noisy. It can be improved by using projective transformations.

Reference

https://learnopencv.com/video-stabilization-using-point-feature-matching-in-opencv/