Road Sign Detection and Recognition using Scale Invariant Feature Transform

Abstract

In a traffic environment, road signs play an important role in regulating the traffic. Recognizing the road signs is a very important task for drivers. An automated system can be built to recognize the road signs, which can support the vehicle drivers on their way and enables automated surveillance. In this research work, a real time method is proposed to detect and recognize the road signs. Computer vision and pattern recognition has played a crucial part in the design of this system. Initially, the captured image from the camera is converted into the HSI (Hue, Saturation and Intensity) colour space and colour segmentation is done by using Bayes classifier. The colour segmentation produces three binary images. Then the morphological operations are used to remove noise in the binary images and the connected components in the binary images are labeled. The method exploits some geometrical properties of road signs to extract the road sign labels from the road scene image. Finally the SIFT method is used to extract the local invariant features of the image labels and recognition is done by matching the extracted features with the stored features of the standard road signs in the database. The MATLAB® implementation of this system handles three different colour (blue, yellow, red) road signs. This system has been trained and tested using the Sri Lankan road signs and the results are effective in detection as well as recognition. 92% of signs are detected and 74% of them are recognized by the system.