Citation

Song JG, Lee JW. Int. J. Automot. Technol. 2023; 24(3): 773-786.

Copyright

(Copyright © 2023, Holtzbrinck Springer Nature Publishing Group)

DOI

10.1007/s12239-023-0064-z

PMID

unavailable

Abstract

Convolutional neural networks (CNNs) have been successful for tasks such as object detection; however, they involve time-consuming processes. Therefore, there are difficulties in applying these CNNs to autonomous driving. Moreover, most autonomous driving technologies require both object detection and distance prediction. However, CNNs that predict distance involve more time-consuming processes than object detection models. In addition, the applications for autonomous driving require object detection and distance prediction accuracy. This paper proposes an end-to-end trainable CNN that can meet these requirements. The proposed CNN accurately implements object detection and distance prediction in real time using stereo images. We demonstrate the superiority of the proposed CNN using stereo images from the KITTI 3D object detection dataset.

Language: en

Keywords

Autonomous driving; Distance prediction; Object detection; Real-time processing