This paper presents a probabilistic vehicle states prediction algorithm by using multi-sensor fusion. The system inputs come in two main varieties: 1) vehicle sensor signal, such as steering angle, longitudinal velocity, longitudinal acceleration and yaw rate and 2) vision sensor signal, such as curvature, slope and distance to lane mark. From these inputs, the algorithm presents the time series prediction of future vehicle states and the corresponding covariance matrixes for the pre-defined future time horizon.
The probabilistic states prediction algorithm consists of two sequential parts. The first part is the estimation part which contains a vehicle filter which estimates current vehicle states and a road filter which approximates the road geometry. The second part is prediction part which consists of a path following model generating future desired yaw rate which acts as a virtual measurement and a vehicle predictor which predicts future vehicle states by maximum likelihood filtering method.
The proposed algorithm has been investigated via test data based closed loop simulation with Smart Cruise Control (SCC) system. Compared to two kind of existing path prediction methods; a fixed yaw rate assumption based method and a lane keeping assumption based method, it has been shown that the states prediction performance can be significantly enhanced by the proposed prediction algorithm. And this enhancement of prediction performance led to capabilities improvement of driver assistance functions of SCC by providing accurate predictions about the future driving environment.