Traffic sign recognition by division of characters and symbols regions

This paper proposes a recognition algorithm of a traffic sign. First, we use a color and shape information in order to extract the candidate of traffic sign. In actual, traffic sign forms the upright rectangular shape. However, the obtained image using the camera mounted in the car can have a traffic sign shape of a rotated rectangular or trapezoid. The region of traffic sign are classified into the closed areas which contain characters and arrows which show the road direction. We generate eight-direction code vectors for an each closed region. Each of input pattern is composed of the eight-direction code vectors. Standard pattern is predefined to the eight-direction code vectors that is generated from closed regions of the characters and arrows without noise. We use DP algorithm to match input pattern to the standard pattern. As the result of DP matching, we get pattern information of an each closed region. DP algorithm is robust at a scale variation of an input pattern when input pattern do not have deficient codes unlike the normal standard codebook. In occasion, input pattern, there are likely to have deficient codes unlike the normal standard codebook. In the case, DP algorithm is not robust at a scale variation of an input pattern. We find the deficient codes of input pattern, and insert them. Therefore, we can get recognition algorithm that is robust at a scale variation of an input pattern. The font of Korean characters in a traffic sign is prescribed to the boldface. Therefore, we apply to this font in experiment. Finally, we show that recognition of characters with various size and symbol is sucessful in experiments.