Systems and Methods for Pedestrian Detection in Images

• US 20140314271A1
• Filed: 04/18/2013
• Published: 10/23/2014
• Est. Priority Date: 04/18/2013
• Status: Active Grant

• Current Assignee: Futurewei Technologies Incorporated (Huawei Investment & Holding Co., Ltd.), Huawei Technologies Co., Ltd. (Huawei Investment & Holding Co., Ltd.)
• Original Assignee: Huawei Technologies Co., Ltd. (Huawei Investment & Holding Co., Ltd.)
• Inventors: Liu, Haibo, Zhang, Philipp, Chen, Wei, Xu, Jingsong, Ni, Zefeng, Wu, Qiang, Zhang, Jian
• Granted Patent: US 9,008,365 B2
• Time in Patent Office: Days
• US Class Current: 382/103
• CPC Class Codes: G06F 18/2411 based on the proximity to a decision surface, e.g. support vector machines
• CPC Class Codes: G06F 18/24323 Tree-organised classifiers
• CPC Class Codes: G06V 10/467 Encoded features or binary features, e.g. local binary patterns [LBP]
• CPC Class Codes: G06V 10/50 by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 40/103 Static body considered as a whole, e.g. static pedestrian or occupant recognition

Abstract

System, apparatus, and method embodiments are provided for detecting the presence of a pedestrian in an image. In an embodiment, a method for determining whether a person is present in an image includes receiving a plurality of images, wherein each image comprises a plurality of pixels and determining a modified center symmetric local binary pattern (MS-LBP) for the plurality of pixels for each image, wherein the MS-LBP is calculated on a gradient magnitude map without using an interpolation process, and wherein a value for each pixel is a gradient magnitude.

Claims

1. A method for determining whether a person is present in an image, comprising:

receiving a plurality of images, wherein each image comprises a plurality of pixels;

determining a modified center symmetric local binary pattern (MS-LBP) feature for the plurality of pixels for each image, wherein the MS-LBP feature is calculated on a gradient magnitude map without using an interpolation process, and wherein a value for each pixel is a gradient magnitude; and

determining that the person is present in one of the images in accordance with the MS-LBP.

2. The method of claim 1, further comprising classifying the image using a linear support vector machine (SVM) that integrates two different features on a single stage classifier, wherein a first of the two different features is created by a first feature descriptor and a second one of the two different features is created by a second feature descriptor, wherein the first feature descriptor is the MS-LBP, and wherein one of the features is the MS-LBP feature created by the MS-LBP.

3. The method of claim 2, wherein one of the two different features comprises a local binary pattern (LBP) feature determined from the plurality of pixels.

4. The method of claim 3, wherein the MS-LBP comprises first vectors and the LBP comprises second vectors and wherein the first and second vectors are concatenated to each other.

5. The method of claim 1, further comprising classifying one of the images using a first layer classifier, wherein no further classifying on the one of the images is performed if the first layer classifier determines that the person is not present in the one of the images.

6. The method of claim 5, further comprising classifying one of the images using a second layer classifier if the one of the images passes the first layer classifier.

7. The method of claim 6, wherein the first layer classifier and the second layer classifier are linked in a non-homogeneous manner.

8. The method of claim 6, wherein the first layer classifier is trained based on the MS-LBP feature.

9. The method of claim 6, wherein the second layer classifier is trained based on a local binary pattern (LBP) feature determined from the plurality of pixels.

10. The method of claim 1, further comprising classifying one of the images as one of containing a pedestrian image and not containing a pedestrian image.

11. The method of claim 10, further comprising training a classifier to classify the images using a plurality of training images.

12. The method of claim 10, further comprising refining a classification using multiple images.

13. The method of claim 12, wherein refining the classification comprises using a tracking-like method.

14. A network component configured for determining a presence of a pedestrian in an image, comprising:

a processor; and

a computer readable storage medium storing programming for execution by the processor, the programming including instructions to;

receive a plurality of images, wherein each image comprises a plurality of pixels;

determine a modified center symmetric local binary pattern (MS-LBP) feature for the plurality of pixels for each image, wherein the MS-LBP feature is calculated on a gradient magnitude map without using an interpolation process, and wherein a value for each pixel is a gradient magnitude; and

determine that the pedestrian is present in one of the images in accordance with the MS-LBP.

15. The network component of claim 14, further comprising instructions to classify the image using a linear support vector machine (SVM) that integrates two different features on a single stage classifier, wherein a first of the two different features is created by a first feature descriptor and a second one of the two different features is created by a second feature descriptor, wherein the first feature descriptor is the MS-LBP, and wherein one of the features is the MS-LBP feature created by the MS-LBP.

16. The network component of claim 15, wherein one of the two different features comprises a local binary pattern (LBP) feature determined from the plurality of pixels.

17. The network component of claim 16, wherein the MS-LBP comprises first vectors and the LBP comprises second vectors and wherein the first and second vectors are concatenated to each other.

18. The network component of claim 15, further comprising instructions to classify one of the images using a first layer classifier, wherein no further classifying on the one of the images is performed if the first layer classifier determines that the pedestrian is not present in the one of the images.

19. The network component of claim 18, further comprising instructions to classify one of the images using a second layer classifier if the one of the images passes the first layer classifier.

20. The network component of claim 19, wherein the first layer classifier and the second layer classifier are linked in a non-homogeneous manner.

21. The network component of claim 19, wherein the first layer classifier is trained based on the MS-LBP feature.

22. The network component of claim 19, wherein the second layer classifier is trained based on a local binary pattern (LBP) feature determined from the plurality of pixels.

23. The network component of claim 14, further comprising instructions to classify one of the images as one of containing a pedestrian image and not containing a pedestrian image.

24. The network component of claim 23, further comprising instructions to train a classifier to classify the images using a plurality of training images.

25. The network component of claim 23, further comprising instructions to refine a classification using multiple images.

26. The network component of claim 25, wherein refining the classification comprises using a tracking-like method.

27. A system for determining whether an image contains an image of a pedestrian, comprising:

a feature computation unit comprising a processor, wherein the feature computation unit is configured to determine a modified center symmetric local binary pattern (MS-LBP) feature for a plurality of pixels for an image, wherein the MS-LBP feature is calculated on a gradient magnitude map without using an interpolation process, and wherein a value for each pixel is a gradient magnitude; and

a classifier/detector configured to determine whether the image contains an image of a person based at least in part on the MS-LBP.

28. The system of claim 27, wherein the classifier/detector is further configured to classify the image using a linear support vector machine (SVM) that integrates two different features on a single stage classifier.

29. The system of claim 28, wherein the two different features comprise the MS-LBP feature and a local binary pattern (LBP) feature determined from the plurality of pixels.

30. The system of claim 27, wherein the classifier/detector is further configured to classify the image using a first layer classifier.

31. The system of claim 30, wherein the classifier/detector is further configured to classify the image using a second layer classifier if the image passes the first layer classifier.

32. The system of claim 31, wherein the first layer classifier and the second layer classifier are linked in a non-homogeneous manner.

33. The system of claim 27, further comprising a pedestrian detection refinement unit configured to refine a classification of the image using multiple images.

34. The system of claim 33, wherein refining the classification comprises using a training-like method.