Maximum Likelihood Multiple Model Filtering for Path Prediction in Intelligent Transportation Systems

摘要

Path prediction is an important step in many automotive applications. The idea is to develop a system which allows a vehicle to predict its own path as well as the path of the vehicles in its vicinity. This helps the driver to acquire an enhanced perception of the road environment. In this paper, we develop a novel filtering method to track the movements of an ego vehicle using measurements from GPS sensors. Vehicle maneuver is captured using different kinematic models. In order to combine the strengths of different models, the proposed filter performs a maximum likelihood selection of model-dependent filter estimates. The proposed filter is called the Maximum Likelihood Multiple Model (MLMM) filter. We show that the MLMM filter can provide sub-meter accuracy in terms of position estimation and works well even when a significant fraction of measurements are missing and with diverse trajectories including those having many curved road segments.