About the T-Rex project
Abstract
The purpose of this project is to create a computer remote controlled car that finds moving objects and follows them. The decisions are made by analyzing video from on board camera. The initial idea to this project came from the DARPA challange, where teams from universities around the world develop self-driving vehicles that try to complete an off-road challange of about 130 miles.
This project gives a good example for using OpenCV library with the .net GUI and threading libraries by using C++ with managed extensions.
This project was developed as a student final project (workshop) by Oded Shahar and Roy Shilkrot from the Academic College of Tel Aviv Yaffo, Under the supervision of Tal Hassner.
We really enjoyed working on this project together with Tal. We would like to commend his attention and willingness to help us with any problem we ran into. Thank you Tal!
Why T-Rex?
Since our project centers on computer vision based on movement analysis, we thought it would be appropriate to name it after an (extict) animal that can see by detecting motion.
In the film Jurassic Park, paleontologist Dr. Alan Grant and a young girl are cornered by a T-Rex. "Don't move," he tells her. "If we don't move he won't see us...". This is in fact not a very accurate assumption at the vision capabilities of the Tyrannosaurus. Scientists think that T-Rex's vision was quite good and there is no reason to think it could only see moving objects. Even so, we think the movie caused a consensus to be formed to believe that Tyrannosaurus-Rex were able to see only moving objects, hence the name.
In this site
In this site you'll find explanation of all of the aspects of the project, from the hardware to the algorithm and testing methods. You will also find the source & compiled code of our debugging and testing platform, with install guides.
Feel free to contact us with any question regarding the project or computer vision.
News Update - 23/8/07
We had another filming day and came up with some very nice results. We found a nice big clear area inside a sports hall that allowed the car to move a long distance.
The videos include both the look from outside and the look from the car camera. The streams are almost totally in sync.
News Update - 19/8/07
We made changes to the way we scatter points on the screen. Now we concentrate the majority of the points in the center of the screen and "neglecting" the borders. This was done according to an assumption that the object is likely to be farther away from the car, and therefore more in the middle of the screen.
We took some more videos of the car going. We failed to achieve a significant success, but the tries came out nice.
You can now also see the view from outside of the car and the view from inside the car.
News Update - 5/8/07
Our latest expiriments yielded nice results.
The car was able to follow to object for a few meters before it lost its "grip".