SYSTEM AND METHOD FOR MODEL BASED PEOPLE COUNTING

• US 20100124357A1
• Filed: 11/17/2008
• Published: 05/20/2010
• Est. Priority Date: 11/17/2008
• Status: Active Grant

• Current Assignee: Kyndryl Incorporated (Kyndryl Holdings, Inc.)
• Original Assignee: International Business Machines Corporation
• Inventors: Hampapur, Arun, Zhai, Yun, Tian, Ying-li
• Granted Patent: US 8,295,545 B2
• Time in Patent Office: Days
• US Class Current: 382/103
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 2207/30196 Human being; Person
• CPC Class Codes: G06T 7/251 involving models
• CPC Class Codes: G06V 20/53 Recognition of crowd images, e.g. recognition of crowd congestion

Abstract

An approach that allows for model based people counting is provided. In one embodiment, there is a generating tool configured to generate a set of person-shape models based on results of a cumulative training process; a detecting tool configured to detect persons in a camera field-of-view by using the set of person-shape models, and a counting tool configured to track detected persons upon crossing by the detected persons of a previously established virtual boundary.

Claims

1. A method for counting people using overhead camera views, said method comprising:

generating a set of person-shape models during a cumulative training process;

detecting persons in a camera field-of-view by using said set of person-shape models, and

counting people by tracking detected persons upon crossing by said detected persons of a previously established virtual boundary.

2. The method according to claim 1, said generating a set of person-shape models further comprising:

computing a quantized gradient map of an input image; and

deriving a probabilistic map of gradient magnitude for each quantized gradient direction by accumulating person-shape models generated by said cumulative training process.

3. The method according to claim 1, said detecting persons in a camera field-of-view further comprising:

constructing at least one gradient map consisting of a plurality of down-sampled input images and a plurality of up-sampled input images, and

convolving a plurality of constructed gradient maps with said set of predetermined person-shape models.

4. The method according to claim 1, said counting people further comprising:

predicting a position of a person in the next frame based on results of the current said person'"'"'s detection and tracking;

performing detection of said person in the spatial proximity of said predicted position;

performing detection of said person in a broader region around said predicted position if no person is detected in said spatial proximity;

updating tracking results by matching detected person with previously tracked persons, and

increasing count upon crossing virtual boundary by said tracked person.

5. A system for counting people using overhead camera views, said system comprising:

at least one processing unit;

memory operably associated with the at least one processing unit;

a generating tool storable in memory and executable by the at least one processing unit, said generating tool configured to generate a set of person-shape models based on results of a cumulative training process;

a detecting tool storable in memory and executable by the at least one processing unit, said detecting tool configured to detect persons in a camera field-of-view by using said set of person-shape models, and

a counting tool storable in memory and executable by the at least one processing unit, said counting tool configured to track detected persons upon crossing by said detected persons of a previously established virtual boundary.

6. The generating tool according to claim 5 further comprising:

a computing component configured to compute a quantized gradient map of an input image, and

deriving component configured to derive a probabilistic map of gradient magnitudes for each quantized gradient direction by accumulating results obtained during the cumulative training process.

7. The detecting tool according to claim 5 further comprising:

a constructing component configured to construct at least one gradient map consisting of a plurality of down-sampled input images and a plurality of up-sampled input images, and

convolving component configured to convolve a plurality of gradient maps with said set of predetermined person-shape models.

8. The counting tool according to claim 5, further comprising:

a predicting component configured to predict a position of a person in the next frame based on results of the current said person'"'"'s detection and tracking;

a detecting component configured to perform detection of said person in the spatial proximity of said predicted position and in a broader region around said predicted position if no person is detected in said spatial proximity;

an updating component configured to update tracking results by matching detected person with previously tracked persons, and

counting component configured to increase count upon crossing virtual boundary by said tracked person

9. A computer-readable medium storing computer instructions, which when executed, enables a computer system to count people using overhead camera views, the computer instructions comprising:

generating a set of person-shape models during a cumulative training process;

detecting persons in a camera field-of-view by using said set of person-shape models, and

counting people by tracking detected persons upon crossing by said detected persons of a previously established virtual boundary.

10. The computer-readable medium according to claim 9 further comprising computer instructions for:

computing a quantized gradient map of an input image; and

deriving a probabilistic map of gradient magnitude for each quantized gradient direction by accumulating person-shape models generated by said cumulative training process.

11. The computer-readable medium according to claim 9 further comprising computer instructions for:

constructing at least one gradient map consisting of a plurality of down-sampled input images and a plurality of up-sampled input images, and

convolving a plurality of constructed gradient maps with said set of predetermined person-shape models.

12. The computer-readable medium according to claim 9 further comprising computer instructions for:

predicting a position of a person in the next frame based on results of the current said person'"'"'s detection and tracking;

performing detection of said person in the spatial proximity of said predicted position;

performing detection of said person in a broader region around said predicted position if no person is detected in said spatial proximity;

updating tracking results by matching detected person with previously tracked persons, and

increasing count upon crossing virtual boundary by said tracked person.

13. A method for deploying a counting tool for counting people using overhead camera views, said method comprising:

providing a computer infrastructure operable to;

generate a set of person-shape models during a cumulative training process;

detect persons in a camera field-of-view by using said set of person-shape models, and

count people by tracking detected persons upon crossing by said detected persons of a previously established virtual boundary.

14. The method according to claim 13, the computer infrastructure further operable to:

compute a quantized gradient map of an input image; and

derive a probabilistic map of gradient magnitude for each quantized gradient direction by accumulating person-shape models generated by said cumulative training process.

15. The method according to claim 13, the computer infrastructure further operable to:

construct at least one gradient map consisting of a plurality of down-sampled input images and a plurality of up-sampled input images, and

convolve a plurality of constructed gradient maps with said set of predetermined person-shape models.

16. The method according to claim 13, the computer infrastructure further operable to:

predict a position of a person in the next frame based on results of the current said person'"'"'s detection and tracking;

perform detection of said person in the spatial proximity of said predicted position;

perform detection of said person in a broader region around said predicted position if no person is detected in said spatial proximity;

update tracking results by matching detected person with previously tracked persons, and

increase count upon crossing virtual boundary by said tracked person.