In automotive tracking applications, using two separate linear state space models for longitudinal and angular movement of objects is a widely applied simplification. The separation is possible if the observed targets are positioned straight ahead and moving in approximately the same direction as the observer, like in adaptive cruise control (ACC) systems. However, in more general scenarios of future tracking applications, object motion may not be limited to certain directions. In inner-city and intersection situations, other road users are passing even perpendicular to the observing vehicle. Most tracking systems of today are not prepared to handle those situations, as the simplified modeling is no longer appropriate. In the paper on hand we will review the commonly used models and state their main drawbacks. The conclusion of these drawbacks is the use of a motion model which reflects a more natural description of typical objects to be considered in automotive applications. All mathematical expressions necessary for an implementation using an extended/unscented Kalman filter are provided. The state space model was designed for radar target tracking but is not limited to radar. With modifications to the measurement equations, the model can be used for camera-based systems as well as for ultrasonic sensors or laserscanner systems.
On motion models for target tracking in automotive applications
Ein Bewegungsmodell zur Zielerfassung bei fahrzeugspezifischen Anwendungen
2001
6 Seiten, 7 Quellen
Aufsatz (Konferenz)
Englisch
Realtime Vision Based Multi-Target-Tracking with Particle Filters in Automotive Applications
British Library Conference Proceedings | 2006
|Observable Dynamics and Coordinate Systems for Automotive Target Tracking
British Library Conference Proceedings | 2009
|