TTraffic Sign Recognition with Hierarchical Convolutional Neural Network

Surucu E. A., DOĞAN H.

26th IEEE Signal Processing and Communications Applications Conference (SIU), İzmir, Türkiye, 2 - 05 Mayıs 2018 identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Cilt numarası:
  • Basıldığı Şehir: İzmir
  • Basıldığı Ülke: Türkiye
  • Anahtar Kelimeler: CNN, hierarchical, traffic signs, deep learning, GTSRB
  • Dokuz Eylül Üniversitesi Adresli: Evet

Özet

Convolutional Neural Network(CNN) becomes one of the most preferred deep learning method because of achieving superior success at solution of important problems of machine learning like pattern recognition, object recognition and classification. With CNN, high performance has been obtained in traffic sign recognition which is important for autonomous vehicles. In this work, two-stage hierarchical CNN structure is proposed. Signs are seperated into 9 main groups at the first stage by using structure similarity index. And then classes of each main group are subclassed with CNNs at the second stage. Performance of the network is measured on 43-classes GTSRB dataset and compared with other methods.