Today, the autonomous vehicles are in the spotlight, the wealthy IT companies and automobile manufacturers are working with great efforts on these smart cars, from which everyone expects that they will revolutionize the 21st century’s personal transportation.
We can read about working prototypes and we know already that these vehicles rely on GPS, radar and computer vision to map their surrounding environment. So, they do not rely on just one system, but decide on the basis of information provided by multiple sensors. From this trio the computer vision is the least reliable.
My master thesis purpose was to create a system which can determine the position of three major traffic signs on a picture. The Stop, Yield and Pedestrian crossing signs were chosen.
My solution is built upon the Local Binary Pattern (LBP) which helped me to create a classifier for each of the three signs. I was curious as how well the newer LBP takes up the challenge against the older Haar feature, which has already been successfully used in object detection.
I have created three image databases, they were necessary for the teaching and testing. The videos and most of the images were taken by me. I have written auxiliary programs that helped automate time-consuming operations.
The next stage was to teach and test the LBP and Haar type classifiers. It was important to find the optimal value of the numerous parameters. I have made both types of classifiers for each of the traffic signs.
With the completion of the last phase, I could carry out the comparative studies and measure the effectiveness of each classifier’s detection.
The last section was the development of a graphical program which could detect the three traffic signs on independent or from video pictures.