Initialization Procedure for Radar Target Tracking without Object Movement Constraints

The tracking of radar targets in automotive applications often relies on certain constraints to the movement of objects. For example, the objects of interest in an ACC system (adaptive cruise control) are other vehicles that are positioned straight ahead and moving in approximately the same direction as the observer. In this case, a single radar measurement (distance to target, bearing angle and Doppler velocity) contains -neglecting measurement noise -full information about the position (by distance and angle) and the movement state. Thus, the initialization of tracks can be done based on a single measurement. Without the mentioned assumption, no information about the movement direction of the object is contained in a single measurement. Theoretically, at least two measurements are necessary to extract information about the object movement direction. But due to severe measurement noise and quantization, even three or more measurements may contain misleading information about the movement state. In this paper we present an initialization procedure for radar target tracking without any constraints to object movement. In the first cycles of a new track, the state estimation is computed by a linear regression method. After that, the track state is handed over to a Kalman filter which does the tracking for the rest of the track's lifetime.