Sensor fusion for vehicle tracking based on the estimated probability

Abstract

The track-to-track fusion (T2TF) algorithm is currently an attractive fusion methodology in the industry because the algorithm can reflect the reliability of sensor tracks. However, the T2TF algorithm cannot be applied when the probability information of the sensor is unknown. The aim of this study is to exploit the T2TF algorithm even in the absence of the probability information of the sensor. The covariance is estimated using the recursive equations of the Kalman filter. In addition, a novel track-association approach using the total similarity is developed to improve association performance. The total similarity complements the defects of the track disposition and the estimated track history. Finally, by fusing the associated tracks using the estimated covariance, the T2TF algorithm is successfully applied to sensors with an unknown covariance. The fusion results are then evaluated using the correct association rate and the optimal subpattern assignment metric. The simulation results obtained show the superiority of the proposed algorithm under three scenarios.