Traffic sign recognition using edge detection and eigen-face: Comparison between with and without color pre-classification based on Hue

Most traffic sign recognition algorithms utilize Template Matching which compare detected sign with templates. Studies on this method have shown outstanding recognition accuracy. Nevertheless, the Template Matching burdens a system in term of memory usage since it has to store numerous templates. Eigen-face is a basic method originated to recognize faces. It is efficient and practical since system only needs to store an Eigenface-image and Weights that associated with it. This paper developed a traffic sign recognition using Eigen-Face algorithm. Instead of using RGB images, the learning was utilized edges. It is more distinctive feature compare to color intensity which varies from yellow, red and blue and additional black symbol. The template signs were first converted into grayscale intensity. Its edges were detected using common Sobel approximation and then concatenated into one matrix. Eigenvalues and Eigenvectors of the matrix's Covariance were then calculated. In this algorithm, the biggest Eigenvector was selected and projected as Eigenface-image. Each traffic sign had unique Weight associated with the Eigenface-image that could be used for recognition. This paper compares how to disperse and distinct each sign's weights with and without color pre-classification based on median of Hue. The recognition with color pre-classification shown clearer weights' distinction between each type of traffic sign yet lower weights' disparity within types.