Object recognition using Haar features and histograms of oriented gradients

• US 8,447,139 B2
• Filed: 04/13/2011
• Issued: 05/21/2013
• Est. Priority Date: 04/13/2010
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Haas, Norman, Li, Ying, Pankanti, Sharathchandra, Guan, Weiguang
• Primary Examiner(s): Allison, Andrae S
• Application Number:
  US13/085,985
• Publication Number:
  US 20110255743A1
• Time in Patent Office: 769 Days
• Field of Search: 382100-107, 382168-172, 382/190, 382/218, 382/224, 382/260, 348/207.11, 348/229.1, 348/221.1, 345/155
• US Class Current: 382/291
• CPC Class Codes: G06F 18/2148 characterised by the process organisation or structure, e.g. boosting cascade
• CPC Class Codes: G06V 10/7747 Organisation of the process, e.g. bagging or boosting
• CPC Class Codes: G06V 20/582 of traffic signs

Abstract

A system and method to detect objects in a digital image. At least one image representing at least one frame of a video sequence is received. A sliding window of different window sizes at different locations is placed in the image. A cascaded classifier including a plurality of increasingly accurate layers is applied to each window size and each location. Each layer includes a plurality of classifiers. An area of the image within a current sliding window is evaluated using one or more weak classifiers in the plurality of classifiers based on at least one of Haar features and Histograms of Oriented Gradients features. An output of each weak classifier is a weak decision as to whether the area of the image includes an instance of an object of a desired object type. A location of the zero or more images associated with the desired object type is identified.

Claims

1. A method for detecting objects in a digital image, the method comprising:

receiving at least one image representing at least one frame of a video sequence comprising zero or more objects of at least one desired object type;

placing a sliding window of different window sizes at different locations in the at least one image;

applying, for each window size and each location, a cascaded classifier comprising a plurality of increasingly accurate layers, each layer comprising a plurality of classifiers;

evaluating, at each layer in the plurality of increasingly accurate layers, an area of the at least one image within a current sliding window using one or more weak classifiers in the plurality of classifiers based on at least one of Haar features and Histograms of Oriented Gradients (HOG) features, wherein an output of each weak classifier is a weak decision as to whether the area of the at least one image within the current sliding window comprises an instance of an object of the desired object type;

identifying, based on the evaluating, a location within the image of the zero or more objects associated with the desired object type; and

training each weak classifier in the plurality of classifiers based on Haar features and HOG features, wherein a selection of a subsequent weak classifier during the training is based on the subsequent weak classifier that provides a strongest separation between desired object types than other available weak classifiers independent of the subsequent weak classifier being associated with one of a Haar feature and a HOG feature.

2. The method of claim 1, further comprising:

visually indicating, in response to the identifying, the location of the zero or more objects.

3. The method of claim 1, wherein the desired object type is one of:

a traffic sign;

a vehicle;

a pedestrian; and

an animal.

4. The method of claim 1, wherein the evaluating further comprises:

combining each decision from all of the classifiers in the plurality of classifiers into a unified decision;

determining that the unified decision indicates that the area of the at least one image comprises an instance of the desired object type;

if another layer exists in the plurality of increasingly accurate layers, evaluating the current sliding window at another layer in the plurality of layers; and

if another layer fails to exist in the plurality of increasingly accurate layers, indicating that the current sliding window represents an instance of the desired object type.

5. The method of claim 1, wherein the evaluating further comprises:

combining each decision from all of the classifiers in the plurality of classifiers into a unified decision;

determining that the unified decision indicates that the area of the at least one image fails to comprise an instance of the desired object type; and

preventing a sliding window currently being evaluated from being evaluated any further.

6. The method of claim 1, further comprising:

transforming the at least one image into a gray-scale image prior to the evaluating.

7. The method of claim 1, wherein the evaluating produces a scalar value for each of the Haar features and HOG features, wherein the decision of each classifier is based on comparing the scalar value to a threshold and an inequality sense.

8. An information processing system for detecting objects in a digital image, the information processing system comprising:

a memory;

a processor communicatively coupled to the memory; and

an object detection system communicatively coupled to the memory and the processor, the object detection system configured to perform a method comprising;

receiving at least one image representing at least one frame of a video sequence comprising zero or more objects of at least one desired object type;

placing a sliding window of different window sizes at different locations in the at least one image;

applying, for each window size and each location, a cascaded classifier comprising a plurality of increasingly accurate layers, each layer comprising a plurality of classifiers;

evaluating, at each layer in the plurality of increasingly accurate layers, an area of the at least one image within a current sliding window using one or more weak classifiers in the plurality of classifiers based on at least one of Haar features and Histograms of Oriented Gradients (HOG) features, wherein an output of each weak classifier is a weak decision as to whether the area of the at least one image within the current sliding window comprises an instance of an object of the desired object type;

identifying, based on the evaluating, a location within the image of the zero or more objects associated with the desired object type; and

training each weak classifier in the plurality of classifiers based on Haar features and HOG features, wherein a selection of a subsequent weak classifier during the training is based on the subsequent weak classifier that provides a strongest separation between desired object types than other available weak classifiers independent of the subsequent weak classifier being associated with one of a Haar feature and a HOG feature.

9. The information processing system of claim 8, wherein the evaluating further comprises:

combining each decision from all of the classifiers in the plurality of classifiers into a unified decision;

determining that the unified decision indicates that the area of the at least one image comprises an instance of the desired object type;

if another layer exists in the plurality of increasingly accurate layers, evaluating the current sliding window at another layer in the plurality of layers; and

if another layer fails to exist in the plurality of increasingly accurate layers, indicating that the current sliding window represents an instance of the desired object type.

10. The information processing system of claim 8, wherein the evaluating further comprises:

combining each decision from all of the classifiers in the plurality of classifiers into a unified decision;

determining that the unified decision indicates that the area of the at least one image fails to comprise an instance of the desired object type; and

preventing a sliding window currently being evaluated from being evaluated any further.

11. The information processing system of claim 8, wherein the method further comprises:

visually indicating, in response to the identifying, the location of the zero or more objects.

12. The information processing system of claim 8, wherein the evaluating produces a scalar value for each of the Haar features and HOG features, wherein the decision of each classifier is based on comparing the scalar value to a threshold and an inequality sense.

13. A computer program product for detecting objects in a digital image, the computer program product comprising:

a non-transitory storage medium readable by a processing circuit and storing instructions for execution by the processing circuit for performing a method comprising;

receiving at least one image representing at least one frame of a video sequence comprising zero or more objects of at least one desired object type;

placing a sliding window of different window sizes at different locations in the at least one image;

applying, for each window size and each location, a cascaded classifier comprising a plurality of increasingly accurate layers, each layer comprising a plurality of classifiers;

evaluating, at each layer in the plurality of increasingly accurate layers, an area of the at least one image within a current sliding window using one or more weak classifiers in the plurality of classifiers based on at least one of Haar features and Histograms of Oriented Gradients (HOG) features, wherein an output of each weak classifier is a weak decision as to whether the area of the at least one image within the current sliding window comprises an instance of an object of the desired object type;

identifying, based on the evaluating, a location within the image of the zero or more objects associated with the desired object type; and

training each weak classifier in the plurality of classifiers based on Haar features and HOG features, wherein a selection of a subsequent weak classifier during the training is based on the subsequent weak classifier that provides a strongest separation between desired target objects than other available weak classifiers independent of the subsequent weak classifier being associated with one or a Haar feature and a HOG feature.

14. The computer program product of claim 13, wherein the desired object type is one of:

a traffic sign;

a vehicle;

a pedestrian; and

an animal.

15. The computer program product of claim 13, wherein the evaluating further comprises:

combining each decision from all of the classifiers in the plurality of classifiers into a unified decision;

determining that the unified decision indicates that the area of the at least one image comprises an instance of the desired object type;

if another layer exists in the plurality of increasingly accurate layers, evaluating the current sliding window at another layer in the plurality of layers; and

if another layer fails to exist in the plurality of increasingly accurate layers, indicating that the current sliding window represents an instance of the desired object type.

16. The computer program product of claim 13, wherein the evaluating further comprises:

combining each decision from all of the classifiers in the plurality of classifiers into a unified decision;

determining that the unified decision indicates that the area of the at least one image fails to comprise an instance of the desired object type; and

preventing a sliding window currently being evaluated from being evaluated any further.

17. The computer program product of claim 13, wherein the method further comprises:

visually indicating, in response to the identifying, the location of the zero or more objects.