FYP Presentation

I'm going to have my final assesment for my undergrad life tommorow... the FYP presentation. I'm a little bit nervous right now-- thanks to my supervisor who said to me: "I think your presentation should be okay- I'm not worried about you". You're not the one worried sir.. I am..

LOL

But thankfully there are a lot of friends who gave me inputs and opinion.. so I do hope I can rehearse and test my slides before the presentation tommorow.. maybe tonight.. after workshop =)

I never shared about my FYP in this blog right? I guess I'll talk about it for a while..

Did you ever wonder what takes the most of our brain to do? It's our vision ability. Or.. at least that's what I heard.. For you people who study biology, please correct me if I'm wrong. So, back to human vision system, did you ever wonder how we are able to process those images we see from our eyes, recognize the difference between an apple and a picture of an apple? Have you ever wondered how we are able to tell easily whether a person in running, walking or jumping? Consider the video of two person dancing below. You can directly recognize the two individuals, their body parts and what they are doing right?

Our vision system is one of the most complicated mysteries to researchers who want to imitate it inside a computer. So far, computer vision technique has been attracting a large number of researchers, with debatably small or mediocre breakthrough.

My project, or you can say, my research is to create a system which is able to process a video sequence of two people interacting with each other, tracking the body parts of each individuals. These information will then be used to cross reference with a database of activites to recognize what activity they are doing...

Sounds simple right? But it was not as simple as that.. Gosh..  It took me a lot of time to grasp the original system, not to mention trying to improve it... But I managed to do it.. somehow.. thanks to His blessings.. =)

The system diagram is shown below..

So basically each frame in the sequence will undergo 5 steps: Background subtraction where the individuals will be extracted, followed by pixel color classification to cluster the pixel based on colors as blobs, Relaxation labeling for simplifying the blobs, then blob tracking and finally, grouping of blobs based on their body parts.. in total I managed to type around 80 pages worth of MATLAB code.. And it will run for fourteen hours when I click the button "execute".

Yes - 14 hours. For an 8 seconds video.

That's why now I'm really amazed with the gift God gave us.. We've taken too many things for granted, but after working on this project I really am humbled. True science and knowledge should bring us closer to our Maker, the Lord our God - not throw us away from Him =)

I managed to improve the original algorithm as well. I did a lot of modification and the errors are down by almost half.. Thanks to the idea of my supervisor and God's guidance =). The original paper is written by S. Park and J. K. Aggarwal, titled ‘Simultaneous tracking of multiple body parts of interacting persons’. Google it if you want. You'll find it.

So to close the post.. let me show you the final output ifI used the earlier video sequence. Different color shows different body parts. It looks crappy, I know, but I'm very happy I got this result even though I know it looks crappy haha..

Let there be light!

By the way.. if you're interested, you can come to my presentation. 5th of May 2009 @TR90 NTU, 14.30 pm.