Fast and robust region of interest extraction for Chinese road signs
Chunsheng Liu, Faliang Chang · 2017
Different types of traffic signs has different colors and shapes located in uncontrolled traffic environments. The detection of different types of traffic signs is a difficult problem in pattern recognition and computer vision. In our study, a region of interest (ROI) extraction method is proposed to extract ROI using color contrast in local regions. We utilize the high contrast in local regions to extract ROIs for Chinese circular prohibition signs and triangular danger signs. Because most popular Chinese signs contain red limit signs and yellow prohibitive signs, this method is designed to extract traffic signs with these two colors. Previous color extraction methods largely rely on threshold in some color channels, which is not robust to color changes. This method utilizes the high contrast in local regions to extract ROIs and is robust to color changes. The experiments demonstrate the ROI extraction is robust to color variability and can save detection time.