Accurate and Efficient Traffic Sign Detection Using Discriminative AdaBoost and Support Vector Regression

Real-time traffic sign detection and recognition has been receiving increasingly more attention in recent years due to the popularity of driver-assistance systems and autonomous vehicles. This paper proposes an accurate and efficient traffic sign detection technique by exploring AdaBoost and support vector regression (SVR) for discriminative detector learning. Different from the reported traffic sign detection techniques, a novel saliency estimation approach is first proposed, where a new saliency model is built based on the traffic sign-specific color, shape, and spatial information. By incorporating the saliency information, enhanced feature pyramids are built to learn an AdaBoost model that detects a set of traffic sign candidates from images. A novel iterative codeword selection algorithm is then designed to generate a discriminative codebook for the representation of sign candidates, as detected by the AdaBoost, and an SVR model is learned to identify the real traffic signs from the detected sign candidates. Experiments on three public data sets show that the proposed traffic sign detection technique is robust and obtains superior accuracy and efficiency.