An efficient real-time speed limit signs recognition based on rotation invariant feature

In this paper, we present a novel visual speed limit signs detection and recognition system. In detection stage, for the purpose of reducing the computational load and further decreasing the error detection rate of speed limit sign, a novel de-noising method based on HOG is presented and apply it to Fast Radial Symmetry Transform approach for circle signs detector. In recognition stage, firstly, a method of Fourier-wavelet descriptor is introduced to extract rotation invariant features which can recognize slant speed limit signs. Then the Support Vector Machines with Binary Tree Architecture are designed to identify categories of signs. Supplementary traffic signs are used to alter the meaning of speed limit signs. We propose an algorithm which is able to recognize supplementary signs with slightly rotated in a region below recognized speed limit signs. Experimental results in different conditions, including sunny, cloudy and rainy weather demonstrate that most speed limit signs and supplementary signs can be correctly detected and recognized with a high accuracy and the average processing time is less then 33ms per frame on a standard 2.8 GHz dual-core PC.