Fast traffic sign detection on greyscale images

This paper describes a traffic sign detection framework for greyscale images. The system is a heterogeneous cascade classifier formed by a rectangle features cascade followed by specific filters for each shape. We define two types of filters: The first one is based on local changes of the gradient direction and the second one is based on the idea of radial symmetry and gives us the centre of circular shapes. The filters use the position and the aspect ratio of the detected features to differentiate between real objects and false alarms of the rectangle cascade. At the end we use the PCA to eliminate the remaining false alarms.