Precision of localization estimates

• US 7,487,056 B2
• Filed: 08/22/2005
• Issued: 02/03/2009
• Est. Priority Date: 03/01/2004
• Status: Expired due to Fees

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Tashev, Ivan
• Primary Examiner(s): Barlow; John
• Assistant Examiner(s): LE, JOHN H
• Application Number:
  US11/210,160
• Publication Number:
  US 20050283328A1
• Time in Patent Office: 1,261 Days
• Field of Search: 702/85, 702/94, 702/95, 702/127, 702/150, 702/188, 342/450, 382/128, 382/239, 382/243, 382/284, 382/103, 414/788.1, 435/6
• US Class Current: 702/94
• CPC Class Codes: G01S 7/00 Details of systems according to groups G01S13/00, G01S15/00, G01S17/00

Abstract

Precision and reliability of localization estimates derived from conventional localization systems are improved through a system and method for post-processing of initial localization data, even in environments which may include noise, reflections, or other interference. Such localization systems include conventional sound source localization (SSL) systems based on microphone array inputs, radio source location systems based on directional antenna array inputs, etc. In general, this post-processing system and method applies statistical real-time clustering to initial localization estimates, and then uses this real-time clustering in a multi-stage process to generate new localization estimates having improved precision and reliability relative to the initial localization estimates.

Claims

1. A system for automatically improving precision of initial localization estimates, comprising using a computing device to perform the following steps:

generating an initial set of localization data comprising a plurality of localization estimates;

dividing a work volume into a predetermined number of overlapping regions;

assigning each localization estimate to any corresponding one or more of the overlapping regions to form one or more clusters of localization estimates in one or more of the overlapping regions;

estimating positions of objects represented by each cluster of localization estimates;

determining whether any of the objects are duplicate objects by comparing the estimated object positions for clusters in overlapping regions;

eliminating each duplicate object; and

providing each remaining estimated object position to populate a set of position estimates, said set of position estimates representing a set of improved localization estimates relative to the initial set of localization estimates, and wherein said set of position estimates is stored for later use by a user.

2. The system of claim 1 further comprising steps for computing a confidence level for each position estimate in the set of position estimates.

3. The system of claim 1 wherein a lifetime is associated with each localization estimate, and further comprising steps for excluding any localization estimate having an expired lifetime from the initial set of localization data.

4. The system of claim 3 further comprising steps for computing the lifetime as a function of estimated object motions.

5. The system of claim 3 wherein each cluster of localization estimates has a separate lifetime, and further comprising steps for computing the lifetime of each cluster as a function of estimated object motions.

6. The system of claim 1 further comprising steps for using the set of position estimates to initialize any subsequent estimating of positions of objects represented by each cluster of localization estimates.

7. The system of claim 1 wherein the number of overlapping regions is user definable.

8. The system of claim 1 wherein the amount of overlap is user definable.

9. The system of claim 1 further comprising steps for using a weight-based threshold to determine whether particular clusters actually represent objects prior to estimating positions of objects represented by each cluster of localization estimates.

10. A computer-implemented process for automatically generating a set of new position estimates from a set of initial position estimates, comprising steps for:

forming one or more clusters of initial position estimates by assigning each initial position estimate in the set of initial position estimates to one or more corresponding overlapping regions covering a work volume being monitored by a receiving array;

computing new position estimates from each cluster of initial position estimates;

comparing the new position estimates from each cluster to determine whether any of the position estimates represent a position of a duplicate object;

estimating new position estimates representing duplicate objects by discarding each new position estimates having a lower computed weight than a corresponding duplicate position estimate; and

providing each remaining new position estimate to populate a set of new position estimates, said set of new position estimates being stored for later use by a user.

11. The computer-implemented process of claim 10 further comprising steps for automatically removing estimates older than a predetermined amount of time from the set of initial position estimates prior to forming clusters of initial position estimates.

12. The computer-implemented process of claim 11 further comprising steps for computing the predetermined amount of time as a function of estimated object motions generated from position estimates over time.

13. The computer-implemented process of claim 10 further comprising steps for computing a confidence level for each new position estimate used to populate the set of new position estimates.

14. The computer-implemented process of claim 10 wherein the number of overlapping regions and amount of overlap is user definable.

15. The computer-implemented process of claim 10 further comprising steps for computing new position estimates from each cluster of initial position estimates only for those clusters having a computed weight which exceeds a predetermined threshold.

16. A method for automatically increasing a reliability and precision of initial localization estimates derived from a receiving array, comprising:

generating a set of initial localization estimates from an input received from a receiving array;

dividing a work volume covered by the receiving array into a set of at least partially overlapping sections;

assigning each initialization estimate to one or more of the overlapping sections to create one or more clusters of initial localization estimates within one or more of the overlapping sections;

identifying any of the overlapping regions wherein clusters of the initial localization estimates represent potential objects;

estimating positions of the potential objects in each overlapping region which includes a cluster representing a potential object;

comparing each estimated position to determine whether any of the estimated positions represents a position of a duplicate potential object;

eliminating estimated positions representing duplicate potential objects; and

providing each remaining estimated position to a set of estimated positions representing new localization estimates having increased reliability and precision relative to the initial localization estimates, said set of estimated positions being stored for later use by a user.

17. The method of claim 16 further comprising computing a confidence level for each estimated position provided to the set of estimated positions.

18. The method of claim 16 wherein a lifetime is associated with each localization estimate, and wherein generating the set of initial localization estimates further comprises excluding any localization estimate whose current age is greater than the associated lifetime.

19. The method of claim 16 wherein the number of overlapping regions and amount of overlap is user definable.

20. The method of claim 16 wherein identifying any of the overlapping regions wherein clusters of the initial localization estimates represent potential objects further comprises:

computing a weight for each cluster;

comparing the weight to a predetermined weight threshold; and

excluding any cluster whose weight is less than the predetermined weight threshold.