One of the main fields in the area of ITS (Intelligent Transportation Systems) is driving assistance systems designed to give the driver as much information as possible about his environment.
One of the main fields in the area of ITS (Intelligent Transportation Systems) is driving assistance systems designed to give the driver as much information as possible about his environment. Part of the vital information needed by the driver is an estimation of his distance from the other vehicles in his driving lane. In this project an attempt was made to implement such a system using a standard video camera and image processing techniques. The developed algorithm could be a first step to a real-time application based on a simple camera and onboard computer.
For an estimation of the distance, the license plate was chosen as a reference object, and the main problem became the identification of the license plate in the video. Due to the nature of the driving environment and the dynamic light conditions, the separation of the license plate from background objects can not be achieved easily.
In order to overcome these problems a formalism of fuzzy maps was introduced. The formalism was applied on color criteria, the Hue and Saturation values of the search area in the frame. Those objects which matched the color based criteria were classified based on their geometric properties. The final classification in each frame was based on a SVM classifier. Inter frame classification was also used, in order to increase system reliability. When possible, motion estimation and tracking techniques were applied, to further increase the chances of correct identification.
From simulation results on a large set of frames, satisfying results were obtained and a good estimation of the distance was observed. A low ratio of false alarms was achieved, and the advantage of the fuzzy maps over simple binarization was shown. Example of the analysis process is shown below.
Step Four: Raw black and white picture after adaptive binarization
Step Five: Object map after geometric features based classification
Step Six: SVM classification results
Step Seven: Final Selection, based on inter frame analysis
The Project was implemented using MATLAB 2006a and 2006b. The films were taken using a digital 3CCD camera, and processed using Adobe Premier and Virtual Dub version 1.6.17. The SVM was created with the OSU-SVM toolbox for MATLAB version 3.00, available under BSD license Here.
A distance estimation system was implemented by image processing tools, and the possibility of such a system was demonstrated. Although real time implementation was not achieved in the MATLAB programming environment, the possibility of such implementation was shown in the project. Further work is needed, both in achieving better filming conditions, in improving the algorithm by the addition of more tests and by real-time implementation.
We are grateful to our project supervisor Mr. Johanan Erez for his help and guidance in this project. We would like to thank Mr. Dori Peleg for his help with the SVM implementation.
We are also grateful to the Ollendorf Minerva Center Fund for supporting this project.