Traffic Sign Detection Based On Color And Shape Features

Traffic sign detection is an important part of Driver Assistance System. The automatic detection of traffic signs is a key component of intelligent vehicles. Traffic signs carry warning or direction information.Neglecting these vital waring or guide information brought by traffic signs is dangerous. Thus how to make the driving safe and convenient has become more and more crucial around the word.Traffic signs are in outdoor environment,thus the detection of traffic signs is heavily influenced by the environment where they are actually located in, such as illumination, shadows, occlusion ect. To reduce these factors effects,traffic signs are usually designed with high-contrasitive color such as red, yellow or blue and special shapes such as triangle or circular. Thus color and shape are important cues for traffic sign detection. In this paper we propose a new traffic sign detection method based on color invariants and pyramid histogram of oriented gradients(PHOG) features. The method works in two stages. First, we extract the color invariants of input image in Gaussian color model, and then segment the image into different regions to get the candidate regions of interests (ROIs) by clustering on the color invariants. Area based filter and symmetry region detection algorithms is adopted to remove the noise regions. Next PHOG is adopted to represent the shape features of ROIs and support vector machine (SVM) is used to identify the traffic signs. The Canny edge operator used to extract object coutour in original PHOG is sensitive to the cluttered background of traffic sign in practical scenarios. Thus, to boost the descriptive power of PHOG, we propose introducing Chromatic-edge to enhance object contour while suppress the noises. Extensive experiments demonstrate that our method can robustly detect traffic signs under varying weather, shadow, occlusion and complex background conditions compared to existed algorithms.