Challenges and Conclusions
Our project was not one of only success, but rather was successful because we had many failures and opportunities to learn. Here's some quick highlights of what we accomplished successfully, and where we had our biggest growth from areas of struggle.
What we did well
Over the course of the semester, we were able to create a working motion amplification program. We learned a lot about DSP tools and theory along the way. Through comparison to existing implementations of motion amplification and countless hours of research and testing, we have created an algorithm that can both identify and amplify small motion in a band range of the users choice. Though this band may look different for each different video, when chosen correctly the result is always a video unmasking micro-motion! This motion is everywhere, from our factories and construction sites to our musical instruments and cradles. With our Eulerian motion amplification algorithm, we can see this movement all around us.
​
Throughout the course of the project, we ran into challenges that changed the way we saw the problem. Though we weren't able to overcome every challenge, each one was an opportunity to learn more about our code worked and about digital signal processing with motion amplification.
​
There were a few rough spots throughout the course of the project. We were unfortunately unable to implement the phase-based amplification technique as we had a hard time conceptually understanding them and the math was a bit above our pay grade. But we did learn a lot about it from reading the papers and attempting to implement it.
​
Since we stuck with the Eulerian amplification method noise was a constant annoyance. It limited how far we could amplify the video before noise dominated and sometimes made the video undecipherable. The Eulerian technique was also quite ram hungry which limited us to only processing short standard definition videos.
Challenges
Future Tasks
In the future we would like to be able to implement the phase-based and approach and be able to process higher resolution video. Finding a faster way to determine the optimal passband for the program to amplify is another task worth working towards.
We also recently saw that a newer paper has been released on the subject so we'll be checking that out.