A multiple hypothesis tracker with interacting feature extraction

The multiple hypothesis tracker (mht) is an optimal tracking method due to the enumeration of all possible measurement-to-track associations. However, its practical implementation is limited by the exponential growth in this enumeration. Feature data, in addition to kinematic measurements, can improve track quality and reduce false tracks. In this paper, a new approach whereby there is an intrinsic interaction between feature extraction and the mht is presented. A measure of the feature extraction quality is input into the measurement-to-track association while the prediction step feeds back information to be used in the next round of feature extraction. The motivation for this interaction between feature extraction and tracking is to improve the performance in both steps. In addition, a track-specific detection probability becomes available to the prior. This prior probability moderates the state covariance growth when measurements are not available for track continuation. This moderation permits growth of the tracking gate when a target is out maneuvering the tracker, while this growth is limited when the target is within the tracking gate but with a low probability of detection. Simulation results demonstrate the benefits of the proposed approach with a small increase in computational load.