Deep Learning Based Intelligent Traffic Volume Measurement


Çelik B., Tortum A.

ICA-EAST 2021, Erzurum, Türkiye, 15 - 17 Aralık 2021, cilt.1, sa.29, ss.343-349

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Cilt numarası: 1
  • Basıldığı Şehir: Erzurum
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.343-349
  • Atatürk Üniversitesi Adresli: Evet

Özet


With the development of Computer Vision technology in recent years, this technology has started to be used in Intelligent Transportation Systems as well as in many different fields. With the use of Computer Vision in Intelligent Transportation Systems, the necessary infrastructure can be provided for real-time intelligent control of traffic signaling times, elimination of traffic delays, and more effective and more dynamic control of traffic conditions. Therefore, in this paper, a traffic volume measurement software system that can detect and count selected vehicle classes is proposed in order to calculate real-time traffic signal intervals. Thus, it is aimed to eliminate time and energy losses caused by traffic conditions. In this study, a software system that can detect two different vehicle classes and perform unique counting according to vehicle class has been established by combining YOLOv3 (You Only Look Once) and SORT (Simple Online and Realtime Tracking) algorithms. The training and testing of the algorithm were carried out with a dataset of 84499 labels in 12652 images. Traffic surveillance videos with different conditions were used to validate the proposed method. When the experimental results were examined, it was determined that the proposed method showed a 92.8% count accuracy rate with high performance.