Track level fusion of extended objects from heterogeneous sensors

The detection and tracking of extended targets is a challenging problem. The accuracy of the detections and the tracking is dependent on the granularity and the quality of the data obtained from the sensors. The systems specialised for automotive environment perception task are mostly a mix of high and low resolution sensors, equipped with their own target state estimation module. The sensors provide the fusion center, the target state estimates along with information on the target shape such as length and width. This paper focuses on deriving fused extended shapes from the set of tracked objects obtained from heterogeneous environment perception sensors. The concept and the procedure described in this paper is unique for the fusion of target extension and shape information. The algorithm is analytically investigated to analyse its performance bounds. A detailed Monte Carlo simulation has been carried out with different trajectories and noise sources to ascertain the performance. The interesting part of this paper is the results obtained using real world sensor data in a real traffic condition using high precision ground truth sensors for the sensor platform and the target vehicle.