Final Project

Problem

Video Stablization

Intuition

Video stabilization is a video processing technique to enhance the quality of input video by removing the undesired camera motions. There are various approaches used for stabilizing the captured videos. Most of the existing methods are either very complex or does not perform well for slow and smooth motion of hand held mobile videos. Hence it is desired to synthesis a new stabilized video sequence, by removing the undesired motion between the successive frames of the hand held mobile video. Various 2D and 3D motion models used for the motion estimation and stabilization.

Approach

The goal of any video stabilization algorithm is to create a new video sequences where the motion between the frames has effectively removed. In general any digital video stabilization algorithm consists of three stages Motion Estimation (ME), Motion Smoother (MS) and Motion Compensation (MC).

2D Methods

The 2D methods in general have low computational complexity and can be solved efficiently. However, 2D methods suffer from potential problems. First, the tracked 2D features can be unreliable due to motion blur and illumination change. Obtaining long feature tracks is also difficult in videos with significant occlusions.

3D Methods

3D methods seek to explore the 3D location of feature points in the scene while calculating the camera pose in 3D space. These works in general handle parallax better than 2D methods, since the motion is physically analyzed in actual 3D space. In these works, the camera path is smoothed and the 3D feature points are reprojected to new camera positions in order to guide the warping or the methods use image-based rendering to synthesize a frame from the original frames. However, the 3D methods suffer from robustness and complexity issues in Structure from Motion.

Evaluation

Evaluates the appearance change between consecutive frames in the results.