The objective of this project is to create a driver assistance system.
Abstract
The objective of this project is to create a driver assistance system. Drivers need to pay attention to a lot of things simultaneously and to estimate a lot of measurements without exact information, which can be a problem if the driver is tired. This system would estimate the distance from the vehicle in front the driver and in principle could alert if the vehicle is to close, thus taking some of the pressure of his hands.
The problem
In order to estimate the distance we should be able to identify some object in the car that we know its properties, thus – knowing its size in the picture ( in pixels ) we could know the distance. It appears that the only object that most of the cars have in common is the license plate. All other features change from design to design.
The problem is actually two problems: finding the license plate and then keeping track of it. The system can be integrated into a more comprehensive driver assistance system.
The solution
Finding the license plate:
In Israel most of the LP are yellow. This is a property of the LP that the system uses to locate it.
How can we identify the color yellow?
It appears that the way we conceive colors has a mathematical model called the CIE space. Using this model the system determines which pixel is yellow. It also appears that license plates have another strong quality that can be estimated using a mathematical object. It what makes the color of objects seem full and “strong”. Using these qualities we can isolate the pixels that belongs to the LP and locate it.
Tracking after the LP :
This part is actually a tracking device without a direct connection the LP problem. Anything can be tracked in a picture. After finding the license plate it seems silly to find it in every frame of the movie, it just takes too much time. The tracking system searches only the surroundings of the location of the license plate in the previous picture, and it uses an auto – correlation function in order to find it.
Tools
The project was developed entirely in MATLAB.
Conclusions
Finding the license plate appears to be a very complicated task – it appears that the colors change drastically when moving from light to shadow and in dependence with the hour. Yellow is not defined in a single valued way. Tracking after the license plate seems to be much more efficient because the processing is done in a gray scale picture and thus is not color dependent. Once there is a “lock” on the license plate the system stays on it during passage through light and shade. This system can be implemented in a more real time environment such as C++ or DSP and connected to the car computer. If connected the information of the speed of the car can be integrated into the system to other purposes.
Acknowledgment
I would like to thank Johanan Erez (our supervisor and the VISL lab engineer) for his full support ideas and understanding.
We are also grateful to the Ollendorff Minerva Center Fund for supporting this project.