HIDDEN MARKOV MODEL FOR CAMERA HANDOFF

• US 20090046153A1
• Filed: 08/13/2007
• Published: 02/19/2009
• Est. Priority Date: 08/13/2007
• Status: Active Grant

• Current Assignee: Fuji Xerox Company Limited (Fujifilm Holdings Corporation)
• Original Assignee: Fuji Xerox Company Limited (Fujifilm Holdings Corporation)
• Inventors: Yang, Tao, Kimber, Donald G., Chen, Francine R.
• Granted Patent: US 8,432,449 B2
• Time in Patent Office: Days
• US Class Current: 348/159
• CPC Class Codes: G06T 2207/10021 Stereoscopic video; Stereoscopic image sequence
• CPC Class Codes: G06T 2207/20076 Probabilistic image processing
• CPC Class Codes: G06T 2207/30196 Human being; Person
• CPC Class Codes: G06T 2207/30232 Surveillance
• CPC Class Codes: G06T 7/292 Multi-camera tracking
• CPC Class Codes: G06V 20/52 Surveillance or monitoring of activities, e.g. for recognising suspicious objects recognising microscopic objects G06V20/69

Abstract

An integrated method for modeling the handoff between cameras for tracking a specific individual, including: creating a representation of overlaps, gaps, and allowable movement among the fields of view of the cameras, wherein the representation is modeled as states in a Hidden Markov Model (HMM); training the HMM using video of people walking through the fields of view of the cameras; selecting a person to be tracked; and identifying the best camera area using the HMM.

Claims

1. An integrated method for modeling the handoff between cameras for tracking a specific individual, comprising:

creating a representation of overlaps, gaps, and allowable movement among fields of view of the cameras, wherein the representation is modeled as states in a Hidden Markov Model (HMM);

training the HMM using video of people walking through the fields of view of the cameras;

selecting a person to be tracked; and

identifying a best camera area using the HMM.

2. The method of claim 1, further comprising identifying a best camera in the best camera area using the HMM.

3. The method of claim 1, wherein Viterbi decoding is used to determine the best path between the set of states representing the camera areas.

4. The method of claim 1, wherein the HMM provides a formal framework for specifying allowable movement between views from multiple cameras.

5. The method of claim 1, wherein the HMM represents the floor layout as viewed through cameras as discrete areas, where an area is represented by a state.

6. The method of claim 5, wherein a hotspot is an area where a person can appear in or disappear from the view of one camera

7. The method of claim 6, wherein each area is either a hotspot, an area not covered by a hotspot, or an area that is not in view of any camera.

8. The method of claim 1, wherein the fields of view as a person moves from one camera to another do not need to overlap or provide complete coverage.

9. The method of claim 1, wherein the model does not require identifying pixel-based correspondence points in multiple cameras.

10. The method of claim 1, wherein blind locations are explicitly modeled in the framework.

11. The method of claim 1, wherein the use of similarity as a feature obviates the need for training the HMM observation model for each new person to be tracked.

12. The method of claim 1, wherein overlap of the region outlining the object with the camera area can be computed at each time sample independently of the identity of the best camera area.

13. The method of claim 1, wherein change in velocity of the person can be incorporated into the model.

14. The method of claim 1, wherein each camera area node can be split into two or more nodes.

15. The method of claim 3, wherein transition probabilities are computed during Viterbi decoding as a function of similarity of velocity from a best region in a previous node to velocity of a best region in a current node.

16. A program of instructions executable by a computer to perform a function for modeling the handoff between cameras for tracking a specific individual, the function comprising the steps of:

creating a representation of overlaps, gaps, and allowable movement among the fields of view of the cameras, wherein the representation is modeled as states in a Hidden Markov Model (HMM);

training the HMM using video of people walking through the fields of view of the cameras;

selecting a Very Important Person (VIP) to be tracked;

identifying the best camera area using the HMM; and

identifying the best camera in the camera area using the HMM.

17. The program of instructions of claim 16, wherein Viterbi decoding is used to determine the best path between the set of states representing the camera areas.

18. The program of instructions of claim 16, wherein the HMM provides a formal framework for specifying allowable movement between views from multiple cameras.

19. The program of instructions of claim 16, wherein the HMM represents the floor layout as viewed through cameras as discrete areas, where an area is represented by a state.

20. The program of instructions of claim 16, wherein each area is either a hotspot where a person can appear or disappear in the view of one camera, an area not covered by a hotspot, or an area that is not in view of any camera.

21. The program of instructions of claim 17, wherein transition probabilities are computed during Viterbi decoding as a function of similarity of velocity from a best region in a previous node to velocity of a best region in a current node.

22. The program of instructions of claim 16, wherein blind locations are explicitly modeled in the framework.