Real-time video tracking system

1. A method for detecting and tracking a target, comprising:

detecting the target using a plurality of feature cues;

fusing the plurality of feature cues to form a set of target hypotheses;

tracking the target based on the set of target hypotheses and a scene context analysis; and

updating the tracking of the target based on a target motion model;

wherein the plurality of feature cues is generated by at least a first detector of a first type and a second detector of a second type, wherein the first and second types are different;

wherein the first detector is a appearance-based detector and the second detector is a silhouette-based detector; and

wherein the tracking is used to compensate for a motion of a radiation source tracked by a radiation imaging device.

2. The method of claim 1 , wherein the first detector and the second detector have substantially the same view of the target.

3. The method of claim 1 , wherein fusing the plurality of feature cues to form a set of target hypotheses comprises fusing a first output from a first detector with a second output from a second detector, wherein the first and second outputs are of different types.

4. The method of claim 3 , wherein the first detector is an appearance-based detector and the second detector is a silhouette-based detector, and the first detector operates on the output of the second detector.

5. The method of claim 1 , wherein the plurality of feature cues is generated by at least a first plurality of detectors, wherein a first view of the target for each detector of the first plurality of detectors is substantially the same, and a second plurality of detectors, wherein a second view of the target for each detector of the second plurality of detectors is substantially the same, wherein the first and second views are different.

6. The method of claim 5 , wherein fusing the plurality of feature cues to form a set of target hypotheses comprises fusing a first output of the first plurality of detectors with a second output of the second plurality of detectors.

7. The method of claim 1 , wherein the scene context analysis comprises monitoring a set of polygons representative of an environment where the target may exist and monitoring at least one scene object within the environment.

8. The method of claim 7 , wherein the set of polygons comprises an entry region where the target may enter the environment or an exit region where the target may exit the environment.

9. The method of claim 1 , wherein tracking the target based on the set of target hypotheses and the scene context analysis comprises initializing tracking, holding tracking, resuming tracking, or terminating tracking of the target based on the scene context analysis.

10. The method of claim 1 , wherein updating the tracking of the target based on the target motion model comprises defining a nonlinear Bayesian tracking problem and solving the nonlinear Bayesian tracking problem with a particle-filter-based tracking algorithm.

11. A method, comprising:

detecting a target with a first plurality of detectors, wherein the first plurality of detectors comprises a first appearance-based detector and a first silhouette-based detector;

detecting the target with a second plurality of detectors, wherein the second plurality of detectors comprises a first appearance-based detector and a second silhouette-based detector;

generating a first plurality of feature cues with the first plurality of detectors;

generating a second plurality of feature cues with the second plurality of detectors;

fusing the first plurality of feature cues and the second plurality of feature cues to create a set of target hypotheses;

tracking the target based on the set of target hypotheses and a scene context analysis; and

updating the tracking of the target based on a motion model;

wherein the tracking is used to compensate for a motion of a radiation source tracked by a radiation imaging device.

12. The method of claim 11 , wherein fusing the first plurality of feature cues and the second plurality of feature cues to create a set of target hypotheses comprises fusing the first plurality of features cues with one another to create a first plurality of target hypotheses, fusing the second plurality of feature cues with one another to create a second plurality of target hypotheses, and fusing the first and second target hypotheses to create the set of target of hypotheses.

13. The method of claim 11 , wherein the first and second appearance-based detectors are video cameras and the first and second silhouette-based detectors are video cameras.

14. The method of claim 11 , wherein the first plurality of detectors has a first view of the target and the second plurality of detectors has a second view of the target, wherein the first and second views are different.

15. The method of claim 11 , wherein the scene context analysis comprises monitoring a set of polygons representative of an environment where the target may exist.

16. A method for detecting and tracking a target, comprising:

detecting the target with a plurality of detectors when the target enters an environment through an entry region, wherein the plurality of detectors is configured to generate a plurality of feature cues, wherein at least two of the feature cues are of different types;

fusing the plurality of feature cues to generate a set of target hypotheses;

tracking the target based on the set of target hypotheses and a scene context analysis of the environment; and

terminating tracking of the target when target exits the environment through an exit region;

wherein the plurality of detectors comprises a first appearance-based detector, a second appearance-based detector, a first silhouette-based detector, and a second silhouette-based detector, wherein the first appearance-based detector and the first silhouette-based detector substantially have a first view of the target, and the second appearance-based detector and the second silhouette-based detector substantially have a second view of the target, wherein the first and second views are different; and

wherein the tracking is used to compensate for a motion of a radiation source tracked by a radiation imaging device.

17. The method of claim 16 , wherein fusing the plurality of feature cues to generate a set of target hypotheses comprises fusing a first plurality of feature cues generated by the first appearance-based detector and the first silhouette-based detector to create a first target hypothesis, fusing a second plurality of feature cues generated by the second appearance-based detector and the second silhouette-based detector to generate a second target hypothesis, and fusing the first and second target hypotheses to create the set of target hypotheses.