Reduced Dimension Lane Detection Method

Detecting road lane is one of the key processes in vision-based driving assistance system and autonomous vehicle system. The main purpose of the lane detection process is to estimate car position relative to the lane so that it can provide a warning to the driver if the car starts departing the lane. This process is useful not only to enhance safe driving but also in self-driving car system. A novel approach to lane detection method using image processing techniques is presented in this research. The method minimizes the complexity of computation by the use of prior knowledge of color, intensity and the shape of the lane marks. By using prior knowledge, the detection process requires only two different analyses which are pixel intensity analysis and color component analysis. The method starts with searching a strong pair of edges along the horizontal line of road image. Once the strong edge is detected the process continues with color analysis on pixels that lie between the edges to check whether the pixels belong to a lane or not. The process is repeated for different positions of horizontal lines covering the road image. The method was successfully tested on selected 20 road images collected from internet.


INTRODUCTION
With the advance of information technology, the use of image processing techniques for advanced driving system has attracted a good number of researchers. One of the earliest comprehensive studies of driver assistance system was presented by Mathias et al in 2007 [14]. The report presented complete hardware and software requirement for advanced driver @ IJTSRD | Available Online @ www.ijtsrd.com | Special Issue Publication | November 2018 Engineering and Technology, Linton University College, Negeri Sembilan, School of Electrical, Electronic and Mechanical Engineering and Information Technology, ITP lane is one of the key processes in based driving assistance system and autonomous vehicle system. The main purpose of the lane detection process is to estimate car position relative to the lane so that it can provide a warning to car starts departing the lane. This process is useful not only to enhance safe driving but driving car system. A novel approach to lane detection method using image processing techniques is presented in this research. The method complexity of computation by the use of prior knowledge of color, intensity and the shape of the lane marks. By using prior knowledge, the detection process requires only two different analyses which are pixel intensity analysis and color is. The method starts with searching a strong pair of edges along the horizontal line of road image. Once the strong edge is detected the process continues with color analysis on pixels that lie between the edges to check whether the pixels belong e or not. The process is repeated for different positions of horizontal lines covering the road image. The method was successfully tested on selected 20

Feature Extraction, Lane Mark, Edge
With the advance of information technology, the use of image processing techniques for advanced driving system has attracted a good number of researchers. One of the earliest comprehensive studies of driver as presented by Mathias et al in 2007 [14]. The report presented complete hardware and software requirement for advanced driver assistance system. In October 2013, a researcher named Yenikaya presented the importance of lane marks detection [5]. This article provides a comprehensive review of vision systems. Yinghua He et al investigate an algorithm that has two modules: boundaries are based on the intensity image, and road areas are subsequently detected based on the full [9]. The combination of these modules can overcome the basic problems due to the inaccuracies in edge detection based on the intensity to the computational complexity of segmentation algorithms based on color images. Experimental results on real road scenes have substantiated the effectiveness of the proposed method. He et al proposed a color-based method for the non road (no lane marking) but the analysis causes computational burden [9]. Tu et al use Hough Transform techniques to extract lane marks from road images. This method requires an analysis of all images pixel and can only detect straight lane marks. For a better approach proposed Line parabolic model [4].
In 2009, a research team led by Lan et al developed a SmartLDWS method to detect lane mainly for smart phone usage. In 2014, Li et al investigated a novel real-time optimal-drivable-region and lane de system for autonomous driving based on the light detection and ranging (LIDAR) and vision data. The system uses a multisensory scheme to cover the most drivable areas in front of a successfully handles both structured and unst roads. The research team also proposed a multi method to handle the problem in producing reliable results [ presented the importance of lane marks detection [5]. This article provides a comprehensive review of vision-based road detection systems. Yinghua He et al investigate an algorithm modules: boundaries are first estimated y image, and road areas are subsequently detected based on the full-color image [9]. The combination of these modules can overcome the basic problems due to the inaccuracies in edge intensity image alone and due complexity of segmentation algorithms based on color images. Experimental results on real road scenes have substantiated the effectiveness of the proposed method. First is the searching mode that searches the lane without any prior information of a road [6]. Second is recognition mode, which is able to reduce the size and change the position of a searching range by predicting the position of a lane through the acquired information in a previous frame. It allows to accurately and efficiently extract the edge candidate points of a lane without any unnecessary searching [6].

Lane Detection Method
Detection can be done by analyzing the whole image [4] or by analyzing the image region by region [6][8].
Color, shape, and texture are three features that are widely used in object detection process. Since most of the objects have different intensity compared to their background, edge feature is used to detect any object's existence in the image. To identify the object we will need additional features that make the object specifically differ from the other objects. Color and shape are attributes that we can use for object identification. Since lane marks are a rectangular white object, as the steps to detect the object then we first need to detect the strong edge and check the color component of the pixel next to the edge. We can calculate the pixel gradient value by using first derivative Gauss mask. The use of first derivative Gaussian mask will result in gradient value of smoothed image [12].

Detection Result
After the algorithm was implemented in Matlab programming language, the method was tested to a good quality road image. We selected good quality because the method was a new method. We also defined the ROI of the image was of 2/3 bottom part of the image. The following figure shows the usage of one of the good quality images to test the proposed method and the first search window line.

Figure2 . Pixel intensity distribution
There were three peaks appearing in the distribution graph, two come from the lane marks which have very high intensity and one peak from the shadow-like object. If we convolve the intensity data with first derivative Gaussian, we will get gradient distribution as shown in the following figure.

Figure3. Gradient distribution
From the gradient distribution graph, we can see there were two high positive gradients. Each high gradient pixel followed by one very high negative gradient. The high positive gradient pixel appears when there was a transition from gray pixels (not-marked road) to white pixels (road lane mark The following result demonstrates how the system can avoid positive false result. The high gradient was detected but the result was no lane detected (no green line).

Figure6. Positive false test result
With this result, we can conclude that the method was accurate enough to detect lane marks. The process repeated for detecting lane marks in the whole image. The set of lane marks position is crucial in providing vehicle behavior to the driver. For visual presentation, the program will change the color of lane marks to green color at the pixels that intersect the marks. The overall detection result is presented in the following figure.

Figure7. Positive false test result
When the detection processes are performed for the whole image each time the line crossing lane marks, the program will automatically record the edge of the detected lane marks.

Conclusion
In this paper, we have presented a novel method for lane marks detection. The method used 1D line as search window to detect lane marks. By using a simple threshold for gradient data and RGB color components, the method can smartly and successfully detect the lane marks. With a small number of pixels to be analyzed for each detection steps and simple mathematical calculation involved in the analysis, this method promises to fulfill one of the criteria for good lane detection method set by Narote et al [13]. For further research, we need to test the effectiveness of the method for real road image captured in different conditions.