Method and apparatus for calibrating sampling operations for an object detection process

• US 7,616,780 B2
• Filed: 09/22/2008
• Issued: 11/10/2009
• Est. Priority Date: 11/24/2004
• Status: Active Grant

• Current Assignee: Adobe Inc.
• Original Assignee: Adobe Systems Incorporated (Adobe Inc.)
• Inventors: Bourdev, Lubomir D.
• Primary Examiner(s): Tabatabai; Abolfazl
• Application Number:
  US12/235,389
• Publication Number:
  US 20090016570A1
• Time in Patent Office: 414 Days
• Field of Search: 382/103, 382/107, 382115-118, 382/227, 382/297, 382/317, 348/77, 348/94, 348/152, 348/154, 348/155, 348169-172, 348/208.16, 348/352, 348/353, 348/356
• US Class Current: 382/103
• CPC Class Codes: G06V 10/255 Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
• CPC Class Codes: G06V 10/75 Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries

Abstract

One embodiment of the present invention provides a system that detects an object in an image. During operation, the system determines a relationship between sampling parameters and a detection rate for an object detection process. The system also determines a relationship between the sampling parameters and a detection speed for the object detection process. The system uses the determined relationships to generate specific sampling parameters. Next, the system performs the object detection process, wherein the object detection process uses the sampling parameters to sample locations in the image. This sampling process is used to refine the search for the object by identifying locations that respond to an object detector and are hence likely to be proximate to an instance of the object.

Claims

1. A method for detecting an object within an image, comprising:

a computer performing;

determining a relationship between sampling parameters and a probability of an object detector detecting within the image an instance of the object;

determining a relationship between the sampling parameters and a detection speed of the object detector, wherein the detection speed of the object detector contributes to an amount of time for an object detection process to complete;

using the determined relationships to generate specific sampling parameters; and

performing the object detection process, wherein the object detection process uses the sampling parameters to sample locations in the image, wherein a sampling process is used to refine the search for an instance of the object within the image by identifying locations that respond to the object detector and are hence likely to be proximate to an instance of the object within the image.

2. The method of claim 1, wherein generating the specific sampling parameters involves balancing detection speed of the object detector against the probability of the object detector detecting within the image an instance of the object.

3. The method of claim 2, wherein generating the specific sampling parameters involves maximizing detection speed of the object detector while maintaining a given minimum probability of the object detector detecting within the image an instance of the object.

4. The method of claim 2, wherein generating the specific sampling parameters involves maximizing the probability of the object detector detecting within the image an instance of the object, while maintaining a given minimum detection speed of the object detector.

5. The method of claim 1, wherein the sampling parameters include:

a sampling function, which determines locations within the image to sample; and

a refinement threshold, wherein if a response obtained while sampling a given location exceeds the refinement threshold, a localized search for the object is performed in the vicinity of the given location.

6. The method of claim 5, wherein the object detection process involves:

using the sampling function to identify locations in the image to sample; and

applying the object detector to the identified locations;

wherein if the object detector'"'"'s response at a given location exceeds the refinement threshold, the method further comprises refining the search by attempting to locate a perfect center for the object in the vicinity of the given location.

7. The method of claim 5, wherein the sampling function is determined based upon empirical results obtained while using the object detector.

8. The method of claim 5, wherein the refinement threshold is determined based upon empirical results obtained while using the object detector.

9. A computer-readable storage medium storing instructions that when executed by a computer cause the computer to perform a method for detecting an object within an image, the method comprising:

determining a relationship between sampling parameters and a probability of an object detector detecting within the image an instance of the object;

determining a relationship between the sampling parameters and a detection speed of the object detector, wherein the detection speed of the object detector contributes to an amount of time for an object detection process to complete;

using the determined relationships to generate specific sampling parameters; and

performing the object detection process, wherein the object detection process uses the sampling parameters to sample locations in the image, wherein a sampling process is used to refine the search for an instance of the object within the image by identifying locations that respond to the object detector and are hence likely to be proximate to an instance of the object within the image.

10. The computer-readable storage medium of claim 9, wherein generating the specific sampling parameters involves balancing detection speed of the object detector against the probability of the object detector detecting within the image an instance of the object.

11. The computer-readable storage medium of claim 10, wherein generating the specific sampling parameters involves maximizing detection speed of the object detector while maintaining a given minimum probability of the object detector detecting within the image an instance of the object.

12. The computer-readable storage medium of claim 10, wherein generating the specific sampling parameters involves maximizing the probability of the object detector detecting within the image an instance of the object, while maintaining a given minimum detection speed of the object detector.

13. The computer-readable storage medium of claim 9, wherein the sampling parameters include:

a sampling function, which determines locations within the image to sample; and

a refinement threshold, wherein if a response obtained while sampling a given location exceeds the refinement threshold, a localized search for the object is performed in the vicinity of the given location.

14. The computer-readable storage medium of claim 13, wherein the object detection process involves:

using the sampling function to identify locations in the image to sample; and

applying the object detector to the identified locations;

wherein if the object detector'"'"'s response at a given location exceeds the refinement threshold, the method further comprises refining the search by attempting to locate a perfect center for the object in the vicinity of the given location.

15. The computer-readable storage medium of claim 13, wherein the sampling function is determined based upon empirical results obtained while using the object detector.

16. The computer-readable storage medium of claim 13, wherein the refinement threshold is determined based upon empirical results obtained while using the object detector.

17. An apparatus that detects an object within an image, comprising:

a determination mechanism configured to,

determine a relationship between sampling parameters and a probability of an object detector detecting within the image an instance of the object, and to

determine a relationship between the sampling parameters and a detection speed of the object detector, wherein the detection speed of the object detector contributes to an amount of time for an object detection process to complete;

a parameter generation mechanism configured to use the determined relationships to generate specific sampling parameters; and

an object detection mechanism configured to perform the object detection process, wherein the object detection process uses the sampling parameters to sample locations in the image, wherein a sampling process is used to refine the search for an instance of the object within the image by identifying locations that respond to the object detector and are hence likely to be proximate to an instance of the object within the image.

18. The apparatus of claim 17, wherein while generating the specific sampling parameters, the parameter generation mechanism is configured to balance detection speed of the object detector against the probability of the object detector detecting within the image an instance of the object.

19. The apparatus of claim 18, wherein the parameter generation mechanism is configured to maximize detection speed of the object detector while maintaining a given minimum probability of the object detector detecting within the image an instance of the object.

20. The apparatus of claim 18, wherein the parameter generation mechanism is configured to maximize the probability of the object detector detecting within the image an instance of the object, while maintaining a given minimum detection speed of the object detector.

21. The apparatus of claim 17, wherein the sampling parameters include:

a sampling function, which determines locations within the image to sample; and

a refinement threshold, wherein if a response obtained while sampling a given location exceeds the refinement threshold, the object detection mechanism is configured to perform a localized search for the object in the vicinity of the given location.

22. The apparatus of claim 21,

wherein the object detection mechanism is configured to,

use the sampling function to identify locations in the image to sample, and to

apply the object detector to the identified locations;

and

wherein if the object detector'"'"'s response at a given location exceeds the refinement threshold, the object detection mechanism is configured to refine the search by attempting to locate a perfect center for the object in the vicinity of the given location.

23. The apparatus of claim 21, wherein the sampling function is determined based upon empirical results obtained while using the object detector.

24. The apparatus of claim 21, wherein the refinement threshold is determined based upon empirical results obtained while using the object detector.