Article Preview
TopThe aim of lane detection and tracking based on image processing is, first, to locate in a sequence frame the lane limits of the road in which the vehicle is engaged and, then, to track these limits in the remaining frames.
The acquired images are, first, pre-processed in order to reduce the noise always caused by the sensors (Min, Liu, & Xu, 2006; Wennan, Qiang, & Hong, 2006; Ming-Dar, et al., 2008), to reduce the execution time based on the processing of a region of interest (Min, Liu, & Xu, 2006; Lim, Seng, Ngo, & Ang, 2009), or also to reduce the perspective effect of the lane by applying the inverse perspective mapping (Juan, Hilario, de la Escalera, & Armingol, 2005; Zu, 2006). The second step extracts the approximate pixels of LM either by detecting their edges (Min, Liu, & Xu, 2006; Ming-Dar, et al., 2008; Chih-Hsein & Chen, 2006; Haiping, Ko, Shil, Kim, & Kim, 2007), or by detecting their regions throw image segmentation (Lim, Seng, Ngo, & Ang, 2009; Serge, Michel, & Xuan, 1989; Chao, Mei, & D, 2010). The third step, the lane detection step, determines effective limits on an acquired image, based on the approximate pixels retained in the second step. Various lane detection methods are proposed in the literature and can be grouped in two categories: model-based approach (Min, Liu, & Xu, 2006; Wennan, Qiang, & Hong, 2006; Ming-Dar, et al., 2008; Romuald, Roland, & Frédéric, 2000) and feature-based approach (Ming-Dar, et al., 2008; Zu, 2006; Craig & Zou, 2007; Jan Pablo & Ozguner, 2000) (see Table 1). Finally, the last step, the lane tracking step allows the continuous detection within all the video sequence by updating, in every frame, the last detected limits. This step is required to minimize the noises and the execution time. Different methods were proposed to carry out this tracking step, they can be broadly classified in two approaches: deterministic-based approach (Liu, Dai, Song, He, & Zhang, 2011; Min, Liu, & Hu, 2006; Yue, Teoha, & Shenb, 2004; Muhammad, Arshad, Irfan, & Yahya, 2007) and stochastic-based approach (Ming-Dar, et al., 2008; M, A, & N, 2007) (see Table 2).