Cuak Giler

Deadline final report untuk 3rd year project tinggal 4 hari sahaja lagi. Takat ni baru ader 20 muka surat. Banyak lagi kena tulis ni.. Adeh pancit. Kenapala menggatal buat tingga kerja sampai last minute? hu hu...

Walk or run? Whatever you do, the computer knows...

Good news. I had managed to sort out some problems in my project. The key problems was how to extract data from the sequence of video taken. There are many different approaches had been done and I had chosen a method based on cyclic motion as shown in a paper published in IEEE journal.

You can imagine your leg as a pendulum. It swings from one back to front and goes back again to form a complete cycle while you are walking or running. As this happens, the distance between your head and the ground and the hips angle changes. This two measurements is enough to allow the computer to determine whether you are walking or running.

Fig. 1 Sequence of walking person


Fig. 2 Sequence of a running person

For my project, I define the initial cycle as when the hip angle is zero. The key characteristic of a walking person is the hip angle is maximum when the head-ground distance is maximum in during the cycle. Whereas for running, both angle and head-ground are maximum. If you wonder why or how this can happen, it is because when you walk, both of your leg have contact with the ground while your legs spread. This causing the distance measured become less. While running, there will be a moment when you are 'on-air' or the proper term is double-float. While flying, you head is in a position bit higher than usual. You may not believe what I said, therefore I will present some results based on my project to illustrate further.

a) Double Support          b) Double Float

Fig. 3 Key difference between walking and running

Once you get the data required, you can used complicated classifier such as Neural Network or simple if else statement to recognise type of motion detected in the video.

Fig. 4 Plot of tracked point and angle hip of a walking person.

My next aim is to understand how to use Neural Network to automatically detect peoples behaviour based on the given data. Hopefully this can be done by this Friday.

The moral of the story, be worried guys! especially those who are living in Big Brother cities. You're being watched! Many surveillance operators nowadays have installed behaviour recognition module in their surveillance system [1, 2] . Obviously the software installed is more advanced and robust than my very simple and crude project. No matter what you do, how fast you walk or when did you left the bag unattended the computer can tell. Be a good citizen ok :-) .

Feature Tracking

Alhamdulillah. I managed to find some useful codes on the net. I am still at early stage. More work to be done. Here are the plot of the guy in the video walking.

Gait Recognition

I would like to invite you to get involved in human gait recognition project. It is for my 3rd year project. I have to develop a program to recognise human walking and running behavior. To get more accurate result result, I need more samples of people walking and running. Here are some sample of walking and running videos.

If you are curious how does it work, let me tell you some basics. If you look closely at the video above, you can see a red line. I put the red line in the video to indicate the lowest position of the persons head while walking or running. As he moves along, you can see his head moves up and down. If we plot the position of this person's head in y-axis against time, you will get a sine-ish curve. From the curve we can find peak and valley. Then we can calculate its period between consecutive peak or valley. The period of running normally less than walking.


So, if you are interested to contribute some video, please email your video to my email. Thanks