Deep Learning Based, Real-Time Object Detection for Autonomous Driving Otonom Araclar icin Derin Ogrenme Tabanli, Gercek Zamanli Nesne Tespiti


Akyol G., Kantarcı A., Celik A. E. , Ak A. C.

28th Signal Processing and Communications Applications Conference, SIU 2020, Gaziantep, Turkey, 5 - 07 October 2020 identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/siu49456.2020.9302500
  • City: Gaziantep
  • Country: Turkey
  • Keywords: Autonomous Driving, Computer Vision, Deep Learning, Kalman Filter, Object Detection
  • Istanbul Technical University Affiliated: Yes

Abstract

© 2020 IEEE.One of the active research topics that maintains its popularity in the field of Computer Vision is the problem of object detection in autonomous cars. Since object detection is a difficult problem, high performance solutions do not work very quickly. Similarly, real-time solutions make compromise on performance. However, due to the nature of autonomous driving, object detection systems must perform in real time and high performance. In this study, Tiny YOLOv3, one of the most successful object detection architectures, was combined with one of the classical object tracking methods, the Kalman filter. A small and real-time object detection system, which increases the model's accuracy without losing its speed, is proposed.