Route guidance sign recognition

We propose a new method that can extract route guidance signs from road scene image taken by an on-board camera and describe its structure. We model the route guidance sign as a two level recursive rectangular model. We represent the rectangle of route guidance sign candidate as a pair of vertical line segments and one or two horizontal lines connected to it. After finding one rectangle we then search for rectangles of the same size aligned horizontally. The character candidate regions are separated from the sign area using the p-tile method, and we represent the character candidate by a circumscribed rectangle. Since the letters are of the same size and placed horizontally, the system can search undetected letters within the horizontal area. The arrow candidate has the largest area of all the character candidates. The system recognizes arrows of three directions (right, left and straight) by template matching. We applied the proposed method to real road scene images, and experimental results showed the effectiveness of this method