METHOD AND APPARATUS FOR DETERMINING A CLASSIFICATION BOUNDARY FOR AN OBJECT CLASSIFIER

• US 20080118161A1
• Filed: 11/14/2007
• Published: 05/22/2008
• Est. Priority Date: 11/17/2006
• Status: Active Grant

• Current Assignee: Alpine Electronics Incorporated (Alps Alpine Co., Ltd.)
• Original Assignee: Alpine Electronics Incorporated (Alps Alpine Co., Ltd.)
• Inventors: Liu, Wei, Yuan, Huai
• Granted Patent: US 8,041,115 B2
• Time in Patent Office: Days
• US Class Current: 382/228
• CPC Class Codes: G06F 18/2411 based on the proximity to a decision surface, e.g. support vector machines
• CPC Class Codes: G06V 2201/08 Detecting or categorising vehicles

Abstract

The invention relates to a method and apparatus for determining a classification boundary between an object, such as a vehicle, and a background, using an object classifier. In an embodiment of the invention, a trained classifier is configured to classify and recognize each a plurality of object images and a plurality of background images. Next, a confidence probability density distribution function is calculated for the vehicle and the background using the determined confidence values for the vehicle images and background images. Once the probability density distribution functions for the vehicle and the background are calculated, the classification boundary between the vehicle and the background is determined using the probability density distribution functions for the vehicle or the background, or both, and a predefined model that is appropriate for the application.

Claims

1. A method for determining a classification boundary between an object and a background, comprising:

recognizing, using a trained classifier, each of a plurality of object images and each of a plurality of background images;

classifying, using the trained classifier, each of the plurality of object images and each of the plurality of background images;

determining a confidence value for each of the plurality of recognized and classified object images and for each of the plurality of recognized and classified background images;

calculating a confidence probability distribution function for an object in the plurality of object images, wherein the confidence probability distribution function for the object in the plurality of object images is based on the confidence values determined for the plurality of object images;

calculating a confidence probability density distribution function for a background in the plurality of background images, wherein the confidence probability density distribution function for the background in the plurality of background images is based on the confidence values determined for the plurality of background images; and

determining a classification boundary between the object in the plurality of object images and the background in the plurality of background images using a predefined model, wherein the predefined model is based on the calculated confidence probability density distribution functions for the object in the plurality of object images or the background in the plurality of background images, or the calculated confidence probability density distribution functions for the object in the plurality of object image and the background in the plurality of background images.

2. The method of claim 1, wherein the predefined model comprises a probability of the object in the plurality of object images and the background in the plurality of background images being incorrectly classified, and the probability meets a predetermined target value.

3. The method of claim 2, wherein the probability of the object in the plurality of object images and the background in the plurality of background images being incorrectly classified that meets the predetermined target value, is calculated by the formula:

min(a∫

_T^∞

f_b(x)dx+b∫

<sub>−

∞</sub>

^Tf_v(x)dx),

wherein min( ) represents a minimization operation, f_v(x) is the confidence probability density distribution function for the object in the plurality of object images, f_b(x) is the confidence probability density distribution function for the background in the plurality of background images, a represents a penalty factor for incorrectly recognizing the background in the plurality of background images, b represents a penalty factor for incorrectly recognizing the object in the plurality of object images, and T represents the classification boundary.

4. The method of claim 3, wherein b is greater than a.

5. The method of claim 1, wherein the predefined model comprises a probability of correctly recognizing the object in the plurality of object images, and the probability meets a predetermined target value.

6. The method of claim 5, wherein the probability of correctly recognizing the object in the plurality of object images that meets the predefined target value, is calculated by the formula:

7. The method of claim 1, wherein the predefined model comprises a probability of incorrectly recognizing the object in the plurality of object images, and the probability meets a predetermined target value.

8. The method of claim 7, wherein the probability of incorrectly recognizing the object in the plurality of object images that meets the predetermined target value, is calculated by the formula:

9. The method of claim 1, wherein the predefined model comprises a probability of correctly recognizing, in a new plurality of images, the object in the plurality of object images, and the probability meets a predetermined target value.

10. The method of claim 9, wherein the probability of correctly recognizing, in the new plurality of images, the object in the plurality of object images that meets the predetermined target value, is calculated by the formula:

11. The method of claim 1, wherein the predefined model comprises a probability of incorrectly recognizing, in a new plurality of images, the background in the plurality of background images, and the probability meets a predetermined target value.

12. The method of claim 11, wherein the probability of incorrectly recognizing, in the new plurality of images, the background in the plurality of background images that meets the predetermined target value, is calculated by the formula:

13. The method of claim 1, wherein the confidence probability density distribution functions for the object and the background are normally distributed.

14. An apparatus for determining a classification boundary between an object and a background, comprising:

a recognizing unit, configured to classify and recognize each of a plurality of object images and background images, using a trained classifier, in order to determine a confidence value for each of the plurality of object images and background images;

a calculating unit, configured to calculate a confidence probability density distribution functions for (1) an object in the object images and (2) a background in the background images, wherein the calculation of the confidence probability density distribution function for the object in the object images is based on the confidence value determined for each object image, and the confidence probability density distribution function for the background in the background images is based on the confidence values determined for each background in the background images; and

a determining unit, configured to determine a classification boundary between the object in the object images and the background in the background images using a predefined model, wherein the predefined model is based on the calculated confidence probability density distribution functions for the object in the object images or the background in the background images, or the calculated confidence probability density distribution functions for both the object in the object images and the background in the background images.

15. The apparatus of claim 14, wherein the predefined model comprises a probability of the object in the object images or background in the background images being incorrectly classified, and the probability meets a predetermined target value.

16. The apparatus of claim 14, wherein the predefined model comprises a probability of correctly recognizing the object in the object images, and the probability meets a predetermined target value.

17. The apparatus of claim 14, wherein the predefined model comprises a probability of incorrectly recognizing the object in the object images, and the probability meets a predetermined target value.

18. The apparatus of claim 14, wherein the predefined model comprises a probability of correctly recognizing, in a new plurality of images, the object in the object images, and the probability meets a predetermined target value.

19. The apparatus of claim 14, wherein the predefined model comprises a probability of incorrectly recognizing, in a new plurality of images, the background in the background images, and the probability meets a predetermined target value.

20. A computer readable medium comprising computer executable instructions adapted to perform the steps of:

recognizing, using a trained classifier, each of a plurality of object images and each of a plurality of background images;

classifying, using the trained classifier, each of the plurality of object images and each of the plurality of background images;

determining a confidence value for each of the plurality of recognized and classified object images and for each of the plurality of recognized and classified background images;

calculating a confidence probability distribution function for an object in the plurality of object images, wherein the confidence probability distribution function for the object in the plurality of object images is based on the confidence values determined for the plurality of object images;

calculating a confidence probability density distribution function for a background in the plurality of background images, wherein the confidence probability density distribution function for the background in the plurality of background images is based on the confidence values determined for the plurality of background images; and

determining a classification boundary between the object in the plurality of object images and the background in the plurality of background images using a predefined model, wherein the predefined model is based on the calculated confidence probability density distribution functions for the object in the plurality of object images or the background in the plurality of background images, or the calculated confidence probability density distribution functions for the object in the plurality of object image and the background in the plurality of background images.