Real-Time Recognition of Blue Traffic Signs Designating Directions

In this research we propose a real-time system to recognize blue traffic signs designating directions. This research is complementary to the previous work done on six annular red signs. The system consists of several processing steps: We firstly label the blue objects in each frame and segment them from the background. After that we try to verify if the segmented blue object is a sign candidate, and then we segment white objects within the blue object. Finally we classify the white objects by matching them to arrow patterns according to geometrical features, or reject them if no arrow pattern is matched. Classification is done using a decision tree. Processing time is about 110 ms/frame, and recognition rate is about 81%.