The research of traffic cones detection based on Haar-like features and adaboost classification

Road recognition plays an important role in the field of unmanned driving. As a common road traffic sign, traffic cones are widely used for indicating temporary roads under certain situations. This paper presents a method to identify traffic cones based on data collected by machine vision system. The real-time captured road image is firstly transformed from RGB color space to HSI color space, and then, the ROI (region of interest) of traffic cone target is extracted by Haar-like feature. Thirdly, four special weak classifiers with Haar-like features and a strong classifier which trained by AdaBoost algorithm are used for recognizing the traffic cones. In this experiment, a strong classifier which composes of 40 weak classifiers is developed, and 9298 positive and negative samples are used for training and calculating the weights of weak classifiers. The experimental results show that 93.4% traffic cones could be successfully detected whereas 6.6% are missing due to overlay in straight road section with long distances away. In addition, the cone detection time in one frame of road image is less than 50ms which indicates the system can meet the detection requirements of the temporary road at low speed.