A Hierarchical Traffic Sign Image Recognition Algorithm Keeping the Local Topology

Traffic sign recognition is crucial in enabling intelligent driving. This paper proposes a robust hierarchical traffic sign image recognition algorithm that maintains local topology. The algorithm is divided into two matching levels: rough matching and fine matching. The multi-scale image's Oriented Fast and Rotated BRIEF (ORB) [1] are extracted in coarse matching. Hamming distance is adopted to measure the similarity of feature points to execute rough and fast matching. In precise matching, the matched candidate feature points are further filtered using the number of features of surrounding points, that is, the Shape Context [2] feature, to achieve precise matching. This hierarchical matching method maintains the local topological characteristics of the image while performing robust and fast matching. In the experiment, the data set provided by Changsha University of Science and Technology was used to assess the performance of the algorithm presented in this article.