Design and implement a tracking system that can track moving vehicles in videos. The system should be able to detect and track multiple vehicles in a video, even if the background also moves (due to airplane motion). You may use any methodology in the literature (or your own) to do this. One method is to compute optical flow, cluster the flow vectors, find and remove the background region, and then the vehicle regions should be left. Another method is to use the interest operators to find the vehicles, kind of like in Video Google. You may use the freely available interest operators (detectors) and descripters. (See URLs.)
We have a large variety of videos available, from very simple ones with a single moving object, to multiple cars and trucks on a simple background, to extremely difficult videos of tanks and trucks in the dessert. Of course, you should start with the simple ones. We have already extracted all the images frames from the videos for your convenience. When you get to our video page http://www.cs.washington.edu/research/VACE/, just click on images. The ones under nima1 are the simplest, and the ones under boeing1 are the most difficult. If you click on videos instead of images, you can get to the actual videos to watch them. However, the nima1 videos are in a strange format that came with a codec that caused our machines to totally hang up. The only way we have found to view the nima1 videos is through using mplayer on Linux. The boeing1 vidoes (which are from the DARPA-sponsored VIVID project) are viewable on most any video player.
Your tracking system should keep track of each vehicle it tracks with identification numbers (1,2,3...) and identify the vehicles on each frame by putting colored boxes around them. Each vehicle gets a different colored box to differentiate (ie, red, green, yellow, etc.) This is shown in the directory called shah for one of the nima videos.
We suggest the following videos for use in this project:
nima1:
boeing1:
Sample Timeline
Extra Credit
Getting the system to work on the more difficult videos will earn extra credit.
URLs
What your report should contain