Motion Amplification
By Thomas Dailey, Trystin Martin, and Jayesh Mate
Overview
The goal of our project was to implement a motion amplification program that takes an input video and produces an output video where the motions are amplified. This was inspired by a video and is based of the work from this MIT paper.
After reading and understanding that paper, we found further research in motion amplification, and decided that it would be our task to implement phased-based video motion processing. See the other paper that inspired us here!
​
​
So what is motion amplification?
Motion amplification commonly comes in two forms. The first of these is called Eulerian motion magnification, or more commonly called linear amplification. This is because all movements passed through a given filter are magnified to show motions not normally seen. However, this has a major drawback in that noise is also amplified. Any noisy elements that exist within the range of the pass-band filter will be magnified to the same amount as actual movement.
​
How do we fix this? The answer is called phase-based motion processing. It uses the principle that the movement of pixels in a video can be characterized by the phase of the Fourier transform of the video frames. After the decomposition of the video into component frequencies, these are then analyzed and amplified accordingly. This method greatly reduces noise characteristics of the video, as noise is processed out by the breakdown and filtering.
Why does this matter?
The applications of micro-motion amplification is extremely varied. A method discussed in the video linked above in our overview is for the analysis of machine motion that is unseen by human eyes. For machines in a factory (or elsewhere), they often run for long periods of time in rhythmic patterns of vibration. This can cause gradual damage to the machine itself, as well as equipment around it. Using this method, the motions of these machines can be detected and analyzed.
​
Another use of this is highlighted as the background above. Motion amplification on parts of the body that perform small movements is extremely useful to physicians. Take, for example, the video of the eye above. Such video can be used to determine the frequency of eye movement, which is often used in the diagnosis of concussions, Parkinson's Disease, and MS.