Extended-Search, Bézier Curve-Based Lane Detection and Reconstruction System for an Intelligent Vehicle

Abstract

To improve the real-time performance and detection rate of a Lane Detection and Reconstruction (LDR) system, an extended-search-based lane detection method and a Bézier curve-based lane reconstruction algorithm are proposed in this paper. The extended-search-based lane detection method is designed to search boundary blocks from the initial position, in an upwards direction and along the lane, with small search areas including continuous search, discontinuous search and bending search in order to detect different lane boundaries. The Bézier curve-based lane reconstruction algorithm is employed to describe a wide range of lane boundary forms with comparatively simple expressions. In addition, two Bézier curves are adopted to reconstruct the lanes' outer boundaries with large curvature variation. The lane detection and reconstruction algorithm -including initial-blocks' determining, extended search, binarization processing and lane boundaries' fitting in different scenarios -is verified in road tests. The results show that this algorithm is robust against different shadows and illumination variations; the average processing time per frame is 13 ms. Significantly, it presents an 88.6% high-detection rate on curved lanes with large or variable curvatures, where the accident rate is higher than that of straight lanes.

Lane Reconstruction for Self-Driving Vehicles on Dynamic Road Networks

“…Huang et al [20] implemented a lane reconstruction system using a pair of Bezier curves. To improve reconstruction quality, they set a threshold and fitted the boundary with the Bezier curves.…”

“…The different types of road markings studied include straight [1], [5], [6], curved-lane [7]- [9], and multi-lanes roads [10]- [13]. Researchers have also studied lane boundary detection and tracking [14]- [16], road markings detection and classification [17], obstacle detection [12], [14], lane departure [18], [19], lane reconstruction [20], [21] and lane type classification [22]. However, previous studies rely on the assumption that the road networks remain static.…”

Novel Motion Planning Method for Mobile Robots Using Velocity Obstacle

“…Different types of splines are usable, e.g. : Bezier [39], B-splines [40], Catmull-Rom [41]. The spline can be divided into segments, and each segment is an degree polynomial.…”

Curve recognition algorithm based on edge point curvature voting

“…The characteristic of this method is that it is highly robust to external interferences (such as utility poles and road signs) in lane line recognition. The models used to fit the lane line primarily include straight-line models, 4–6 segmented line models, 7,8 spline curve models, 9–14 and hyperbolic models. 15–17 Various studies 4–6 have indicated that the straight-line model is relatively easy to utilize, and its lane line fitting speed is faster than others.…”

“…Chu et al 7 and Duong et al 8 adopted the segmented straight-line model to fit the curve lane line; their model can describe the direction of the front road but cannot accurately describe the bending degree of the lane. In the literature, 10–13 the Bezier curve model, which is a spline curve model, was applied to fit curved lane lines. When the Bezier curve model is used to fit a lane line with a large curvature, some control points may be taken from areas outside the lane line, which is quite difficult.…”