A Lane Assessment Method Using Visual Information Based on a Dynamic Bayesian Network

The perception and the interpretation of the vehicle's surrounding are the core modules in any advanced driving assistance system, and the first two stages of every autonomous driving solution. The purpose of the perception stage is to assess the state of the relevant objects (objects' positions, shape, class, orientation, speed) using the primary sensorial information. The purpose of the interpretation stage is to use the objects assessment in order to achieve situation assessment, that is, identification of the instantaneous relations between the traffic environment elements as well as their evolution in time. The work presented in this article lies within the interpretation layer; our aim is to identify on which of the road's lanes is the ego-vehicle currently traveling. We refer to the own vehicle as the ego-vehicle and to its current driving lane as the ego-lane. The ego-vehicle is equipped with a stereovision perception system, and the proposed solution is to identify the ego-lane number by matching the visually detected lane landmarks with the corresponding map landmarks, available from an original extended digital map. Additionally, the visually detected vehicles are used for ego-lane number identification. The proposed solution is a probabilistic one, in the form of a Bayesian network, which uses as evidence the visual cues provided by the stereovision perception system. In order to incorporate the dynamic nature of the problem, a temporal dimension was added to the network, resulting in a dynamic Bayesian model. Experimental results illustrate the efficiency of the proposed method in complex situations, and the improved efficiency of the dynamic approach over the static one.