Search

This article presents a two-dimensional (2-D) color, shape, and texture-based stop sign recognition algorithm. The theoretical base of establishing a suitable 2-D correlation coefficient threshold value that can both eliminate nonstop sign objects and minimize the possibility of false-positive identifications is presented. Both the one-dimensional (1-D) algorithm, which has been developed by the authors, and the 2-D algorithm provide a simple and feasible means to automatically extract stop signs from roadway inventory video images with a very competitive computation speed of less than 25 ms for processing a 400 × 300 pixel image. The 2-D algorithm was able to correctly detect all of the images that were correctly detected by the 1-D algorithm. In addition, the 2-D algorithm could correctly detect two of the remaining five images that could not be correctly detected by the 1-D algorithm. The type of images that can be correctly detected by the 2-D algorithm but not the 1-D algorithm is significant because there is a need to recognize the tilted stop signs. The 2-D algorithm can detect stop signs tilted up to 35°. Similarly, the 2-D algorithm can detect a stop sign with its surface being blocked up to 20%. The improved detectability, with only a marginally increased time in the image processing, suggests that the 2-D algorithm is a better algorithm than the 1-D algorithm for processing stop sign images from video logging of roadway inventory. This algorithm can be extended to recognize other signs, such as yield signs, with a slight modification. [ABSTRACT FROM AUTHOR]