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

The research in the field of Advanced Driving Assistance Systems (ADAS) is oriented towards improving the safety, quality, mobility, and efficiency of driving. The first generation of ADAS was limited to mere passive assistance through warnings; for such systems the perception of the environment through various, complementary sensors was the main objective. As the ADAS evolve and incorporate active assistance, the importance of the proper and coherent traffic scene understanding has also increased. In the driving context, the situation assessment refers to the comprehension of the current state of the traffic environment elements: the ego-vehicle, the road, the driver and the other road users. While the perception provides objects assessment (reconstruction of the sensorial detected objects in the scene: objects’ position, form, classification, speed, orientation etc.), the interpretation is responsible for assessing the instantaneous relation between the previously mentioned objects, as well as their evolution in time. Hence, there is an important connection between the perception and the interpretation of the environment: the outputs of the perception modules constitute the inputs for interpretation module. Due to the noisy and uncertain nature of the sensorial information, and due to the fact that not all quantities can be directly measured at all time instances, most approaches for situation assessment are based on probabilistic models. In this paper, we propose a solution for identifying the ego-vehicle travelling lane (egolane number) using the objects assessment of the surrounding traffic objects, detected by the stereovision based perception system. The lane identification is done by analyzing the relationships between the objects in the traffic scene (ego-vehicle, lane markings, and other vehicles), which according to (Rendon-Velez, Horváth, & Opiyo, 2009) is the definition of situation assessment. Therefore, this task is referred to as lane assessment in the rest of the paper. The main idea of the solution is to use both static and dynamic environment objects and their relationship with the ego-vehicle in order to infer A LANE ASSESSMENT METHOD USING VISUAL INFORMATION BASED ON DYNAMIC BAYESIAN NETWORK