Device and method for detecting targets in images based on user-defined classifiers

• US 20110051999A1
• Filed: 03/28/2008
• Published: 03/03/2011
• Est. Priority Date: 08/31/2007
• Status: Active Grant

• Current Assignee: Leidos Innovations Technology, Inc.
• Original Assignee: Lockheed Martin Corporation (Martin Marietta Corporation)
• Inventors: Krahnstoever, Nils O., Tu, Peter H., Doretto, Gianfranco, Rittscher, Jens
• Granted Patent: US 8,295,543 B2
• Time in Patent Office: Days
• US Class Current: 382/103
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06F 18/254 of classification results, e.g. of results related to same input data
• CPC Class Codes: G06F 18/40 Software arrangements specially adapted for pattern recognition, e.g. user interfaces or toolboxes therefor
• CPC Class Codes: G06V 10/245 by locating a pattern; Special marks for positioning
• CPC Class Codes: G06V 10/774 Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06V 10/809 of classification results, e.g. where the classifiers operate on the same input data
• CPC Class Codes: G06V 10/945 User interactive design; Environments; Toolboxes

Abstract

A device and method for detecting targets of interest in an image, such as people or objects of a certain type. Targets are detected based on an optimized strong classifier descriptor that can be based on a combination of weak classifier descriptors. The weak classifier descriptors can include a user-defined weak classifier descriptor that is defined by a user to represent a shape or appearance attribute that is characteristic of parts of the target of interest. The strong classifier descriptor can be optimized by selecting a subset of weak classifier descriptors that exhibit improved performance in detecting targets in training images.

Claims

1. A target detection device for detecting predetermined targets that are captured within an image, the device including:

a target region identification device that is configured to identify a region of interest in the captured image, the region of interest including a target having geometries consistent with the predetermined targets; and

a classification device that is configured to determine whether the region of interest includes the predetermined targets by analyzing the region of interest with a strong classifier descriptor that describes at least one of shape attributes and appearance attributes that are characteristic of the predetermined targets,

wherein the strong classifier descriptor is based on a subset of weak classifier descriptors that is selected from a plurality of weak classifier descriptors that each describe at least one of shape attributes and appearance attributes that are characteristic of parts of the predetermined targets, and wherein the subset of weak classifier descriptors includes at least one user-defined weak classifier descriptor that is defined by a user based on database images including the predetermined targets.

2. The target detection device according to claim 1, further including an image input device that is configured to input an image.

3. The target detection device according to claim 1, wherein the predetermined targets are at least one of people and objects.

4. The target detection device according to claim 1, wherein the subset of weak classifier descriptors is selected from the plurality of weak classifier descriptors by iteratively selecting the subset of weak classifier descriptors that exhibits improved performance in detecting the predetermined targets.

5. The target detection device according to claim 4, wherein, the subset of weak classifier descriptors is selected from the plurality of weak classifier descriptors by determining the subset of weak classifier descriptors that exhibits the lowest error rate in detecting the predetermined targets.

6. The target detection device according to claim 4, wherein the subset of weak classifier descriptors is selected from the plurality of weak classifier descriptors by iteratively selecting the subset of weak classifier descriptors using an Adaboost algorithm.

7. The target detection device according to claim 1, wherein the strong classifier descriptor is adapted to a camera site by iteratively selecting the subset of weak classifier descriptors that exhibits improved performance in detecting the predetermined targets from training images taken from the camera site.

8. The target detection device according to claim 1, wherein the strong classifier descriptor is adapted to an environmental condition by iteratively selecting the subset of weak classifier descriptors that exhibits improved performance in detecting the predetermined targets from training images taken with the environmental condition.

9. The target detection device according to claim 1, wherein the at least one user-defined weak classifier descriptor is defined from contours of parts of the predetermined targets in the database image.

10. The target detection device according to claim 9, wherein the classification device is configured to determine whether the region of interest includes a vertical response and a horizontal response that are within a predetermined threshold of the contours from the user-defined weak classifier descriptor.

11. The target detection device according to claim 1, wherein the at least one user-defined weak classifier descriptor is defined based on region differences of the predetermined targets in the database image.

12. The target detection device according to claim 11, wherein the user-defined weak classifier is defined as a difference in image intensity values between regions of the predetermined targets, and wherein the classification device is configured to determine whether the region of interest includes a first region and a second region having a difference in intensity values that are within a predetermined threshold of the user-defined weak classifier descriptor.

13. The target detection device according to claim 1, wherein the strong classifier descriptor is based on a linear weighted combination of the subset of weak classifier descriptors.

14. A tracking system for tracking detected targets over a multi-camera network, including the target detection device according to claim 1, and a plurality of cameras.

15. A method of detecting predetermined targets that are captured within an image, the method including:

identifying a target region of interest in the captured image, the target region of interest including a target with geometries consistent with the predetermined targets; and

determining whether the region of interest includes the predetermined targets by analyzing the region of interest with a strong classifier descriptor that describes at least one of shape attributes and appearance attributes that are characteristic of the predetermined targets;

wherein the strong classifier descriptor is based on a subset of weak classifier descriptors that is selected from a plurality of weak classifier descriptors that each describe at least one of shape attributes and appearance attributes that are characteristic of parts of the predetermined targets, and wherein the subset of weak classifier descriptors includes at least one user-defined weak classifier descriptor that is defined by a user based on database images including the predetermined targets.

16. The method according to claim 15, wherein the subset of weak classifier descriptors is selected from the plurality of weak classifier descriptors by iteratively selecting the subset of weak classifier descriptors that exhibits improved performance in detecting the predetermined targets.

17. The method according to claim 15, wherein, the subset of weak classifier descriptors is selected from the plurality of weak classifier descriptors by determining the subset of weak classifier descriptors that exhibits the lowest error rate in detecting the predetermined targets.

18. The method according to claim 15, wherein the subset of weak classifier descriptors is selected from the plurality of weak classifier descriptors by iteratively selecting the subset of weak classifier descriptors using an Adaboost algorithm.

19. A method of making a target detection device for detecting predetermined targets that are captured within an image, the method including:

generating a plurality of weak classifier descriptors that describe at least one of shape attributes and appearance attributes that are characteristic of parts of the predetermined targets, including generating at least one user-defined weak classifier descriptor that is defined by a user based on database images including the predetermined targets;

determining the performance of each of the plurality of weak classifier descriptors by analyzing the accuracy of each weak classifier descriptor in detecting the predetermined targets from a training image; and

constructing a strong classifier descriptor from a subset of the plurality of weak classifier descriptors based on the performance of each weak classifier descriptor, wherein the subset includes at least one user-defined weak classifier descriptor.

20. The method according to claim 19, wherein the strong classifier descriptor is constructed from the subset of the plurality of weak classifier descriptors by determining the subset of weak classifier descriptors that exhibits the lowest error rate in detecting the predetermined targets.

21. The method according to claim 19, wherein strong classifier descriptor is constructed from the subset of the plurality of weak classifier descriptors by iteratively selecting the subset of weak classifier descriptors using an Adaboost algorithm.

22. The method according to claim 19, wherein the strong classifier descriptor is adapted to a camera site by iteratively selecting the subset of weak classifier descriptors that exhibits improved performance in detecting the predetermined targets from training images taken from the camera site.