I am currently working as a research intern in Bangalore pursuing my bachelor thesis. I have been here for the past 2 months working with the Computer Vision Group on a project they had started .Its been a great experience so far and I felt I should update the blog with the work that I’ve been doing.
An important part of my project is to track multiple users with good amount of accuracy. I have tried using median flow tracker with good results.The main problem has been the fact that fast/random motions tend to destroy the box. So I have used Haar Cascades to detect faces and then use the resulting Box as the input to Median Flow tracker. And anytime the box gets destroyed , haar cascade is run on the image again and MFT runs afresh on the new Box.
I have also added multiple user tracking( currently only 2 people can be tracked ).The project can be downloaded from https://github.com/ritsz/MFTracker_CPP . Have a look at the readme.txt before running the code. The 51st line of the code might have to be changed depending upon the place you have haarcascade_frontalface_alt.xml in you file system.
Here is a video of the results I had when I ran the code myself.(http://www.youtube.com/watch?v=UhfeqD2q9gc&feature=plcp).
Currently,as an improvement I am trying to implement the Condensation Algorithm on the points the MFT returns. The results seem to be better than what I had when I used Kalman Filter. The results of the tracking with that are here. http://www.youtube.com/watch?v=f6xASxysFv8&feature=youtu.be .
Lot of work still to be done. I am currently learning the OpenNI framework to work with Asus Xtion Pro Live ( A barebone version of the Kinect developed specifically for Robotics application). I want to port all these into kinect. Kinect itself has awesome tracking capabilities but it too is lacking in a few areas(mainly the fact that it uses only depth image to track a human or what it thinks is a human). The tracking mechanism has to be integrated with the project that is going on in research institute.I guess all of this should keep me busy for a while. 🙂