In recent years, in order to minimize traffic accidents, developing driving assistance systems for security has attracted much attention. Lane detection is an essential element of avoiding accidents and enhancing driving security. In this chapter, the authors implement a novel real-time lighting-invariant lane departure warning system. The proposed methodology works well in different lighting conditions, such as in poor conditions. The experimental results and accuracy evaluation indicates the efficiency of the system proposed for lane detection. The correct detection rate averages 97% and exceeds 95.6% in poor conditions. Furthermore, the entire process has only 29 ms per frame.
TopIntroduction
Driving Assistance Systems (DAS) increases safe and secure driving. This system used to adjust, enhance, and automate the driving. The majority of traffic accidents happen because of drivers lack attention. Driving Assistance Systems reduces the driver workload and provides security. The system either alerts the driver whenever a dangerous situation is encountered. National Highway Transportation Safety Administration (NHTSA) state that a large percentage of accidents caused by distracted drivers and unintended lane departures(1,575,000 accidents annually)(Kumar & Simon, 2015). Lane Departure Warning Systems (LDWS) is an important module in Intelligent Transportation Systems. LDWS based on monocular vision, see itself as a key to avoiding deaths by accident with high reliability and low cost. Different systems implemented in order to identify the road lane markings and the departure condition on the road. These systems can be arranged into two approaches, model-based and feature-based (McCall et al., 2006). McCall et al. (2006) implemented VioLET system which used steerable filters to do lane marking detection. Others propose Standard Hough Transform (STH) (Son et al.,2015; Deng & Han, 2013), Inverse Perspective Mapping (IPM) (Deng & Han, 2013; Li et al., 2014; Aly, 2008), RANSAC (Deng & Han, 2013;Aly, 2008; Guo et al., 2015), spline fitting (Son et al., 2015; Aly, 2008), Catmull-Rom splines (Guo et al., 2010), or a clustering method (Son et al., 2015). However, the major problem of these techniques is unsatisfactory performance and high computational complexities under various lighting conditions. This research focused on a vision-based application, the performance for road lane detection is superior (McCall et al., 2006; Aly, 2008; Son et al., 2015; Borkar & Smith, 2009). In this work, we present a novel road lane detection markings for lane departure warning system under various lighting in real-time, which works in daytime, rainy and at nighttime. The proposed system can be detected to both curved as well as the straight road in different weather conditions. In this research, the major contributions of our work can be summarized as follows:
- •
Vanishing point detection used to extract the region of interest (ROI) is an important task of pre-processing step (Son et al., 2015), aiming at reducing the computational complexity due to the processing time. Processing entire pixels of the full image is unnecessary.
- •
Otsu method (Otsu, 1979) is a key operation to segment candidate lines in lane detection (Li et al., 2014; Borkar & Smith, 2009). Thus, we applied the Otsu threshold method to improve our algorithm and deal with the lighting problem.
- •
Standard Hough Transform (SHT)(Duda & Hart, 1972) has certain disadvantages such as its large false positive rate and the calculation complexity (Son et al., 2015). Then, in this paper, a variant of Standard Hough Transform is used in order to solve this problem. Progressive Probabilistic Hough Transform (PPHT)(Matas, 2000; Mammeri et al., 2014) an effective approach in terms of decreasing the false positive rate as well as reducing the amount of computation necessary to extract road lane markings by using the difference in the fraction of votes necessary to extract road lane accurately with different numbers of supporting pixels.