Probabilistic and Holistic Prediction of Vehicle States Using Sensor Fusion for Application to Integrated Vehicle Safety Systems

This paper presents a probabilistic and holistic prediction algorithm for vehicle states using multisensor fushion. Three concerns are mainly considered in this paper, i.e, reliable and reasonable information fusion, extension of predicted states, and real-time evaluation of prediction uncertainties. The main idea of this paper is that a state-prediction problem can be solved as a multistage optimal estimation problem based on the current vehicle motion, a road geometry description in the current body-fixed frame, a path-following behavior model, and the error covariance of each. The prediction algorithm consists of two sequential parts. The first part is estimation, which contains a vehicle filter that estimates the current vehicle states, and a road geometry filter, which approximates the road geometry. The second part is prediction, which consists of a path-following model that generates the future desired yaw rate, which acts as a virtual measurement, and a vehicle predictor, which predicts the future vehicle states by a maximum-likelihood filtering method. The prediction performance of the proposed method has been investigated via vehicle tests. Moreover, its applicability to integrated vehicle safety system (IVSS) has been validated via computer simulation studies. It is shown that the state-prediction performance can be significantly enhanced by the proposed prediction algorithm compared with conventional methods. The enhancement of the prediction performance allows for the improvement of driver assistance functions of an IVSS by providing accurate predictions about the future driving environment.