DNet-CNet: A novel cascaded deep network for real-time lane detection and classification

Zhang, Lu; Jiang, Fengling; Yang, Jing; Kong, Bin; Hussain, Amir; Gogate, Mandar; Dashtipour, Kia

Abstract

Robust understanding of the lane position and type is essential for changing lanes in autonomous vehicles. However, accomplishing this task in real time with high level of precision is not trivial. In this paper, we propose a novel cascaded deep neural network (DNet-CNet) for real-time end-to-end lane detection (DNet) and classification (CNet). The proposed model can simultaneously predict the lanes position and types. DNet integrates the spatial features extracted from the encoder with those from the decoder to compensate for the lower dimensional encoded data and edge information. Furthermore, the output of DNet is fused with the input image for real time lightweight lane classification model (CNet). The combined features exploit the inherent colors and shape of lanes to improve classification accuracy. Experimental results on the benchmark TuSimple, Caltech-lanes and ELAS datasets show that, the model proposed achieves superior lane detection and classification accuracy in real-time as compared to Cascade-CNN.

Citation

Zhang, L., Jiang, F., Yang, J., Kong, B., Hussain, A., Gogate, M., & Dashtipour, K. (2023). DNet-CNet: A novel cascaded deep network for real-time lane detection and classification. Journal of Ambient Intelligence and Humanized Computing, 14, 10745-10760. https://doi.org/10.1007/s12652-022-04346-2

• Journal Article Type: Article
• Acceptance Date: Jul 11, 2022
• Online Publication Date: Jul 30, 2022
• Publication Date: 2023-08
• Deposit Date: May 21, 2024
• Journal: Journal of Ambient Intelligence and Humanized Computing
• Print ISSN: 1868-5137
• Electronic ISSN: 1868-5145
• Publisher: Springer
• Peer Reviewed: Peer Reviewed
• Volume: 14
• Pages: 10745-10760
• DOI: https://doi.org/10.1007/s12652-022-04346-2
• Keywords: Lane, Detection, Classification, Cascade, CNNs
• Public URL: http://researchrepository.napier.ac.uk/Output/3609110