A deep learning based fast lane detection approach. (February 2022)

• Record Type: Journal Article
• Authors: Oğuz, Erkan; Küçükmanisa, Ayhan; Duvar, Ramazan; Urhan, Oğuzhan
• Is Part Of: Chaos, solitons and fractals. Volume 155(2022)
• Publication Date: 2022-02
• Subjects: ADAS -- Lane detection -- Deep learning -- Real-time; Chaotic behavior in systems -- Periodicals; Solitons -- Periodicals; Fractals -- Periodicals; Chaotic behavior in systems; Fractals; Solitons; Periodicals; 003.7
• DOI: 10.1016/j.chaos.2021.111722
• Languages: English
• ISSNs: 0960-0779

Abstract

Highlights: Novel 1-D signal based lane detection approach. Real-time lane detection capability with 34 fps. Significantly faster than the previous deep learning based approaches by maintaining the lane detection performance at similar level. Suitable for embedded platforms with its very low computational load. Abstract: Autonomous vehicles have recently been very popular and it seems to be causing a major transformation in the automotive industry. A vital component for autonomous vehicles is lane keeping systems. The performance of lane keeping systems is directly related to the lane detection accuracy. For lane detection, various sensors are commonly used. In this paper, a vision based robust lane detection system using a novel 1-dimensional deep learning approach is proposed. Challenging situations as rain, shadow, and illumination reduces the overall performance of vision based approaches. Experimental results show that the performance of proposed approach outperforms existing approaches in literature including these challenging situations in terms of detection performance versus processing speed assessment. Although deep learning based methods that provide high performance have difficulties on low-capacity embedded platforms, the proposed method stands out as a solution with its significantly lower processing time.