Unscented Kalman Filter Design for Automotive Safety Applications
Manolis Tsogas, Aris Polychronopoulos, Angelos Amditis
ICCS, Greece
8th International Conference on Information Fusion, Philadelphia, USA
25 - 29 July 2005
Abstract
Filtering is one of the most fundamental tasks related with target tracking. Especially, in applications such as advanced driver assistant systems, where the estimation of the state of the detected objects – humans, moving and stationary obstacles, is crucial for the immediate warning of the driver. In such systems, due to the motion of the vehicle hosting the sensing devices, the motion modeling of the tracked objects cannot be implemented sufficiently using only a simple linear motion model; thus, more complicated nonlinear models are preferred. In order to be achieved, the extended Kalman filter is used, where the first-order linearization of the non linear system is used for the propagation of the Gaussian random variable which approximates the state distribution of the detected object. Nevertheless, this can introduce large errors in the state estimation and in some cases can lead to the divergence of the filter. In order to avoid sub-optimal performance, the unscented Kalman filter can be chosen, while a new curvilinear model is applied which takes into account both the turn rate of the detected object and its tangential acceleration, leading to a more accurate modeling of its movement. The performance of the UKF using the proposed model in the case of automotive applications is proven to be superior compared to the performance of the extended Kalman filter or to the conventional linear filter because it allows effective tracking of the preceding vehicles especially in the case where they perform complex maneuvers such as turning or lane changing while accelerating. In this paper, a vehicle tracker is developed using the unscented Kalman filter utilizing the proposed curvilinear model and is compared with the extended and the linear Kalman filters in the case of predefined simulated scenarios.
For more information, please contact Angelos Amditis for LATERAL SAFE or Aris Polychronopoulos for ProFusion 2
|
