A Global Motion Model for Target Tracking in Automotive Applications

In common automotive radar tracking systems, simple linear models are used to track targets separately in longitudinal and angular direction relative to the own vehicle (or sensor) position. Under the special condition that the observed targets are straight ahead and moving nearly in the same direction as the observing vehicle, like in adaptive cruise control (ACC) systems, those models work well. In more general scenarios, where movements of other vehicles have to be tracked in all possible directions and all around the vehicle (e.g. in inner-city or intersection situations), the modeling is insufficient. In this paper we review the drawbacks of the commonly used models and present a more general motion model for automotive tracking systems. All necessary expressions for an implementation using an extended or unscented Kalman filter are given. Even if designed for radar systems, the state model is not limited to a special type of sensor. It can be used for ultrasonic or laser scanner systems as well as for vision-based systems with a different measurement model.