Modern tracking systems are required to estimate the distribution oftargets in a surveyed scene while coping with a variety of uncertainties,such as unknown number of targets, unknown target motion models,missed detections and clutter (false measurements). An even more chal-lenging tracking scenario may occur when trying to track targets in ascene with Field-of-View (FOV) constraints. Such situations may arisein urban environments including man-made occlusions (e.g., buildings),or in environments involving topographic constraints (e.g., mountains).In such situations the tracking system needs to reason about informationnot readily accessible, when the target resides in unobserved regions ofthe environment.We address the problem of tracking a single, non-maneuvering tar-get, moving through a scene containing unobserved regions that are apriori known to the tracker. A modied Probabilistic Data Association(PDA) lter is developed, that can take into account unobserved re-gions, in addition to target originated measurements, missed detectionsand clutter. The modication is based on a `negative' information con-cept, whereby expected but actually missing measurements in a scenecontaining unobserved regions may be regarded as useful informationthat can be exploited by the tracking system. In order to use this kindof information it is formulated as a ctitious measurement that embod-ies the essence of the negative information. The ctitious measurement,and its associated measurement noise covariance, are formulated basedon the unobserved region's geometry. This measurement is used to up-date the target state estimate and error covariance by using a regularKalman ltering approach within the standard PDA framework.The performance of the modied PDA lter is studied via numeri-cal simulations. The simulations demonstrate track continuity when thetarget resides in occluded regions, with a controlled growth of the asso-ciated error covariance, thus facilitating robust tracking when the targetbecomes detectable again.
展开▼