Real-time traffic sign recognition from video by class-specific discriminative features

  • Authors:

  • Andrzej Ruta;Yongmin Li;Xiaohui Liu

  • Affiliations:

  • School of Information Systems, Computing and Mathematics, Brunel University, Uxbridge, Middlesex UB8 3PH, UK;School of Information Systems, Computing and Mathematics, Brunel University, Uxbridge, Middlesex UB8 3PH, UK;School of Information Systems, Computing and Mathematics, Brunel University, Uxbridge, Middlesex UB8 3PH, UK

  • Venue:
  • Pattern Recognition
  • Year:
  • 2010

Quantified Score

Hi-index 0.01

Visualization

Abstract

In this paper we address the problem of traffic sign recognition. Novel image representation and discriminative feature selection algorithms are utilised in a traditional three-stage framework involving detection, tracking and recognition. The detector captures instances of equiangular polygons in the scene which is first appropriately filtered to extract the relevant colour information and establish the regions of interest. The tracker predicts the position and the scale of the detected sign candidate over time to reduce computation. The classifier compares a discrete-colour image of the observed sign with the model images with respect to the class-specific sets of discriminative local regions. They are learned off-line from the idealised template sign images, in accordance with the principle of one-vs-all dissimilarity maximisation. This dissimilarity is defined based on the so-called Colour Distance Transform which enables robust discrete-colour image comparisons. It is shown that compared to the well-established feature selection techniques, such as Principal Component Analysis or AdaBoost, our approach offers a more adequate description of signs and involves effortless training. Upon this description we have managed to build an efficient road sign recognition system which, based on a conventional nearest neighbour classifier and a simple temporal integration scheme, demonstrates a competitive performance in the experiments involving real traffic video.