A comparative study on street sign detection
Yang Liu, Xiaojin Gong, Liu Ji-lin · 2012
In order to seek for robust features to describe the street signs, a machine learning based comparative study is proposed. The extraction of descriptors is divided into five steps, including pre-processing, image transformation, block designing, local feature computation and normalization. Several detectors are built using the linear support vector machine by considering the information of color, gradients and texture. The evaluation of them is discussed in detail Moreover, we propose our own street sign dataset since the lack of public ones, and make a statistical analysis on it. Experiments show that different information contributes diversely to the detection performances when adopting different feature computation methods. And the detectors built by robust features can detect street signs with excellent achievements.