Lane change detection algorithm on real world driving for arbitrary road infrastructures

摘要

This study presents a novel investigation of a recently developed model-based approach estimating the lane position of a vehicle circulating within divided road infrastructure. The suggested methodology makes smart use of the spatiotemporal information provided by the embedded sensor technology in the automobile related to object detection in the vehicle neighborhood as well to road infrastructure. Contrary to other similar works no real time monitoring of the entire vehicle environment or high definition images frequently associated with expensive and complex procedures are required. Furthermore, no integration of additional devices is considered while low cost computations are involved. The proposed closed loop decision scheme exploits only a minimal number of measurements available by the vehicle equipment. Different lane recognition criteria based on lane identification calculations and/or lane change detections are conceived. At any time a prioritization selection strategy intelligently defines the most appropriate criterion to be employed depending upon the present context. Real time observations are filtered taking into consideration both the related reliability level and the requirements of the employed lane level decision structure. The currently available information may be associated with a previously taken decision in order to determine the vehicle level lane position. Performance metrics are obtained using real trajectories from vehicles running in motorway stretches A1 and A3 in Paris region. It is shown that the proposed approach provides satisfactory lane detection estimation while a sensitivity is measured when not reliable information is available. Micro statistical analysis explains the algorithm behavior when noisy data while future improvements are also discussed.