Lane detection algorithm for an onboard camera

Abstract

After analysing the major causes of injuries and death on roads, it is understandable that one of the main goals in the automotive industry is to increase vehicle safety. The European project SPARC (Secure Propulsion using Advanced Redundant Control) is developing the next generation of trucks that will fulfil these aims. The main technologies that will be used in the SPARC project to achieve the desiderated level of safety will be presented. In order to avoid accidents in critical situations, it is necessary to have a representation of the environment of the vehicle. Thus, several solutions using different sensors will be described and analysed. Particularly, a division of this project aims to integrate cameras in automotive vehicles to increase security and prevent driver's mistakes. Indeed, with this vision platform it would be possible to extract the position of the lane with respect to the vehicle, and thus, help the driver to follow the optimal trajectory. A definition of lane is proposed, and a lane detection algorithm is presented. In order to improve the detection, several criteria are explained and detailed. Regrettably, such an embedded camera is subject to the vibration of the truck, and the resulting sequence of images is difficult to analyse. Thus, we present different solutions to stabilize the images and particularly a new approach developed by the "Laboratoire de Production Microtechnique". Indeed, it was demonstrated in previous works that the presence of noise can be used, through a phenomenon called Stochastic Resonance. Thus, instead of decreasing the influence of noise in industrial applications, which has non negligible costs, it is perhaps interesting to use this phenomenon to reveal some useful information, such as for example the contour of the objects and lanes.

Lane identification and path planning for autonomous mobile robots

“…found that an entropy-based edge extraction method was valuable due to its ability to adapt to varying scenes [2]. Real time lane recognition was successfully implemented using a modified Otsu algorithm on the histogram of the image in [3], and a stochastic resonance method was able to use noise caused by vehicle vibrations to help detect contours of objects and lanes in [4]. In general, these methods are too time consuming to implement and test and thus beyond the scope of many IGVC competitors.…”

Design of lane keeping assist system for autonomous vehicles

“…A predefined or adaptive percentage of the image can be used to determine the size of region of interest [16]. Besides, some researchers divide the image horizontally [26] or vertically [27] into small parts. By conducting different strategies for rural and urban areas, Jeong and Nedevschi [28] applied predefined percentage to split ROIs for rural ways and adaptive percentage for highways.…”

Calibration of an embedded camera for driver-assistant systems

“…To be able to avoid building a system that works only in specific situations, which will not fulfill the SPARC objectives, the algorithm will implement several approaches to detect the desired lane. Thus, as described in [3], the algorithm will use multiple hypotheses of detection which will track multiple models of lanes. This method will then fuse the results of all the tested models, and provide a solution with a higher level of confidence.…”

“…Referring to appendix I, we use equations (8), (9) and (10), in order to get the coordinates of B = x p|w , y p|w , z p|w . Then, it is obvious that B should be expressed in the coordinate frame of the calibration plane, which can be easily done with (3). Finally, the information in the coordinate system of the camera (O(x |c ; y |c ; z |c )) can be found by using the previous calibration technique (see section III).…”

“…Moreover, it is easy to have the position of the camera C = x C|w , y C|w , z C|w , which is an extrinsic parameter of the calibration procedure. As the points B and C are naturally expressed in the coordinate system of the calibration plane, we will transform R |w in R |p by using the equation (3). Applying this transformation, the point R |p will let appear the angles α and β which are the unknown.…”