Method for combining boosted classifiers for efficient multi-class object detection

• US 7,769,228 B2
• Filed: 04/21/2005
• Issued: 08/03/2010
• Est. Priority Date: 05/10/2004
• Status: Active Grant

• Current Assignee: IP3 2017, Series 200 of Allied Security Trust I (Allied Security Trust)
• Original Assignee: Siemens AG, Siemens Corp. (Siemens AG)
• Inventors: Bahlmann, Claus, Kohler, Thorsten, Pellkofer, Martin, Zhu, Ying, Comaniciu, Dorin
• Primary Examiner(s): Bella; Matthew C
• Assistant Examiner(s): Rosario; Dennis
• Application Number:
  US11/111,640
• Publication Number:
  US 20050249401A1
• Time in Patent Office: 1,930 Days
• Field of Search: 706/39, 706/14, 706/43, 382155-161, 382/220, 382/224, 382/165
• US Class Current: 382/155
• CPC Class Codes: G06F 18/211 Selection of the most significant subset of features
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06V 10/771 Feature selection, e.g. selecting representative features from a multi-dimensional feature space
• CPC Class Codes: G06V 10/7747 Organisation of the process, e.g. bagging or boosting

Abstract

A method for training a system for detecting multi-class objects in an image or a video sequence is described. A common ensemble of weak classifiers for a set of object classes is identified. For each object class, a separate weighting scheme is adapted for the ensemble of weak classifiers. A method for detecting objects of multiple classes in an image or a video sequence is also disclosed. Each class is assigned a detector that is implemented by a weighted combination of weak classifiers such that all of the detectors are based on a common ensemble of weak classifiers. Then weights are individually set for each class.

Claims

1. A computer implemented method for training a system for detecting multi-class objects in an image or a video sequence, the method performed by the computer comprising the steps of:

identifying an ensemble of weak binary classifiers that are common to a set of object classes, wherein each object class shares a common set of features with the other object classes, and wherein each feature of said common set of features is associated with a weak binary classifier in said ensemble of weak binary classifiers,

for each object class, forming an individual strong classifier by adapting a separate set of weights for the ensemble of weak classifiers, wherein a same ensemble of weak binary classifiers is incorporated into each individual strong classifier by the separate set of weights for each object class,

wherein said steps of identifying an ensemble of weak binary classifiers and, for each object class, forming an individual strong classifier are performed by a computer processor.

2. The method of claim 1 wherein one of the object classes is a distinguished class and the ensemble of weak classifiers are selected from the distinguished class.

3. The method of claim 2 wherein the distinguished class has a high occurrence of at least one object in the image or the video sequence.

4. The method of claim 1 wherein the ensemble of weak classifiers is learned from a data set comprising all of the object classes.

5. The method of claim 4 wherein the weights for each object class are adjusted with respect to the common ensemble of weak classifiers and positive training examples in each object class.

6. The method of claim 1 wherein the ensemble of weak classifiers is a combination of weak classifier ensembles, each of which is individually learned from a single object class.

7. The method of claim 6 wherein the combination is a mathematical union of the ensembles of weak classifiers.

8. The method of claim 6 wherein the combination is a mathematical intersection of the ensembles of weak classifiers.

9. The method of claim 6 wherein the combination of weak classifier ensembles is a subset determined by voting.

10. The method of claim 1, further comprising posterior re-training of the set of weights for each object class wherein a misclassification error is reduced.