Models for tracking in automotive safety systems

Abstract: This thesis treats the three main components of a tracking system: motion models, sensor models and implementation architecture. A tracking system utilizes models to extract as much information as possible from a set of sensors, with the purpose of estimating the state of a dynamic system. In other words, it constitutes a foundation for sensor data fusion in an automotive safety system. Two modeling frameworks are presented in this thesis, concerning vehicle motion and usage of radar detections in tracking. The motion model takes the control input from the driver into aspect with increased prediction accuracy as result. Uncertainties in driver style can be formally handled and maneuver classification is possible using a multiple model filter. Information such as object orientation can be extracted from radar data if multiple detections from each object are available and can be accurately modeled. A tracking filter making use of such sensor models is introduced and it is shown how the complex data association problem can be facilitated by joining similar hypotheses into groups. Apart from model design, several practical questions arise when a system is to be implemented in a real-time environment. Such issues are discussed and a general framework for designing and implementing a real-time tracking system is presented. We conclude that a tracking system, using modern estimation techniques, is well suited for sensor data fusion in an automotive environment. Keywords: Automotive, vehicle tracking, radar detections, motion models, active safety, preventive safety.