Abstract—This paper addresses the problem of finding the host vehicle’s lateral position on a multi-lane road, using information obtained by processing video sequences. A very important cue for lane identification is the class of the boundaries of the current lane. This paper presents a reliable solution for lane boundary type identification, based on frequency analysis of the gray level profile of these boundaries, assuming that the current lane is already detected. The lane boundary information is combined with the obstacle information, through a Bayesian Network which will output, frame by frame, the probability of the vehicle to be positioned on each lane of the road. The probability result will be propagated throughout the sequence by a Particle Filter. I. INTRODUCTION dvanced Driving Assistance Systems can significantly improve the driving experience, while also increasing the overall traffic safety. An important prerequisite for any ADAS action is the proper assessment of the situation of the host vehicle and of the surrounding traffic. Part of this situation assessment is the knowledge about the host vehicle’s position on the road. There are several systems that can help us to gain this knowledge: satellite navigation systems can provide a rough position estimate, inertial systems can fill in the gaps of satellite positioning (and filter the estimates), and lane detection systems can tell us the position within a lane. With the help of a map, we can infer an approximate position on the road, or at least we can tell we are on the side of the road corresponding to our direction of driving. Unfortunately, when we have multiple lanes for a driving direction, the problem is not that simple: the navigation systems are not precise enough to tell us on what lane we are, the lane detection systems may not detect all lanes, due to occlusions from other vehicles, and the direction of our driving does not help. In the literature, there exist several approaches for accurately positioning the host vehicle on the road, and estimating the lane on which the host vehicle is travelling on. Manuscript received January 27, 2012. This paper was supported by the project "Doctoral studies in engineering sciences for developing the knowledge based society-SIDOC” contract no. POSDRU/88/1.5/S/60078, project co- funded from European Social Fund through Sectorial Operational Program Human Resources 2007-2013, and by the and by the POSDRU-EXCEL post- doctoral program, financing contract POSDRU/89/1.5/S/62557. Voichita Popescu, Radu Danescu and Sergiu Nedevschi are with the Technical University of Cluj-Napoca, Computer Science Department (e-mail: {firstname.lastname}@cs.utcluj.ro). Department address: Computer Science Department, Str. Memorandumului nr. 28, Cluj-Napoca, Romania. Phone: +40 264 401484. Generally, the solutions are based on GPS localization which is then enhanced by additional vehicle and/or on-board sensors such as inertial navigation systems, odometers, vision sensors, inter-vehicle communication systems, digital maps. Different vision enhanced lane level positioning systems are proposed in [1], [2], [3], [4]; these methods also use detailed digital maps of the environment. A method for lane level positioning based on inter-vehicle communication is presented in [5]. This paper proposed an original solution for lane level positing on a multi-lane road, based on an on-board stereovision processing system and an extended digital map [6]. The contributions of this paper are a novel method for lane boundaries classification and an original method based on a Bayesian Network (BN) for lane estimation. The network is used for correlating the visual information with the map information; additionally, the information about other vehicles is used in the network for lane estimation. The frame by frame results are tracked using a particle filter in order to take into consideration the time evolving nature of the problem. In this approach, roads with three to six lanes per driving direction are considered. The solution is designed for structured roads (roads with marked lane boundaries). Fig. 1 illustrates the overview of the proposed solution for on-road position estimation. The system contains four functional blocks. The first block delivers 3D information through stereo image processing. The second block consists of the tracking and classification of the obstacles and of the lane boundaries; this block provides the evidence for the third block in the architecture. The third block is the probabilistic reasoning block; it performs frame by frame reasoning using a Bayesian network [7], [8] approach. The forth block performs a temporal filtering (tracking) of the instantaneous beliefs provided by the static On-Road Position Estimation by Probabilistic Integration of Visual Cues Voichita Popescu, Radu Danescu, Sergiu Nedevschi A Fig. 1 Solution overview for on-road position estimation 2012 Intelligent Vehicles Symposium Alcalá de Henares, Spain, June 3-7, 2012 978-1-4673-2118-1/$31.00 ©2012 IEEE 583