Template-based target object detection in an image

• US 9,349,076 B1
• Filed: 12/20/2013
• Issued: 05/24/2016
• Est. Priority Date: 12/20/2013
• Status: Expired due to Fees

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Liu, Jiongxin, You, Qian, Jackson, Mark Alan, Pang, Hoi-Cheung
• Primary Examiner(s): Lefkowitz, Sumati
• Assistant Examiner(s): Potts, Ryan P
• Application Number:
  US14/137,686
• Time in Patent Office: 886 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06F 18/24 Classification techniques
• CPC Class Codes: G06Q 30/06 Buying, selling or leasing transactions
• CPC Class Codes: G06V 10/751 Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
• CPC Class Codes: G06V 40/103 Static body considered as a whole, e.g. static pedestrian or occupant recognition

Abstract

Systems and methods are provided for detecting target object(s) within image(s) based on selective template matching. More specifically, the systems and methods relate to template generation, selection and matching based on the identification of regions of interest within image(s). Training images showing target object(s) can be obtained and regions of interest that are deemed more likely to contain part(s) of the target object can be identified based on the training images. Subsequent to the identification of regions of interest, templates for target object detection can be generated based thereon. Templates can be applied on testing images. Based on the test application of templates, a subgroup of templates can be selected to serve as a basis for target object detection in subsequent images.

Claims

1. A computer-implemented method for mannequin detection within an image, the computer-implemented method comprising:

under control of a hardware computing device configured with specific computer-executable instructions,

obtaining a first set of electronic training images depicting mannequins, wherein each electronic training image of the first set includes a first target object for detection, wherein the first target object corresponds to a detectable portion of a mannequin;

identifying a plurality of regions within each electronic training image of the first set, wherein a region of the plurality of regions includes at least a portion of the first target object for detection;

for each region of the plurality of regions, determining an associated likelihood of the region showing at least part of the first target object based, at least in part, on a location of the region relative to a corresponding electronic training image in accordance with a location distribution of the first target object over the first set of electronic training images and on a measure of coverage by an unobstructed portion of the first target object relative to the area of the region;

generating a set of templates based at least in part on a subset of the plurality of regions, wherein the previously determined associated likelihood of each region of the subset exceeds a threshold value defining a high likelihood;

causing configuration of a first classifier for detecting a second target object, wherein the second target object corresponds to a detectable portion of a mannequin, and wherein the configuration of the first classifier is based at least in part on a subset of the set of templates;

causing application of the subset of templates on a second set of electronic training images depicting mannequins to detect the second target object from the second set of electronic training images using the first classifier; and

causing configuration of a second classifier for detecting a third target object from a target set of electronic images, wherein the configuration of the second classifier is based at least in part on a template of the applied subset of templates and wherein a depiction of mannequins in the target set of electronic images is unknown.

2. The computer-implemented method of claim 1, wherein the first, second, and third target objects are a same target object.

3. The computer-implemented method of claim 1, wherein the region of the plurality of regions depicts at least an obstructed portion of the first target object.

4. The computer-implemented method of claim 1, wherein identifying the plurality of regions comprises labeling a plurality of areas within each electronic image of the first set as containing at least part of the first target object.

5. The computer-implemented method of claim 1, wherein the threshold value is predetermined.

6. A computer-implemented method comprising:

under control of a hardware computing device configured with specific computer-executable instructions,

for each region of a plurality of regions within an electronic image, determining an associated likelihood of the region showing at least part of a first target object based, at least in part, on a location of the region relative to the electronic image in accordance with a location distribution of the first target object over a first plurality of electronic images and on a measure of coverage by an unobstructed portion of the first target object relative to the area of the region, wherein the electronic image belongs to first plurality of electronic images;

selecting a region of interest from the plurality of regions, wherein the associated likelihood of the region of interest exceeds a threshold value defining a high likelihood;

generating a plurality of templates based at least in part on the region of interest; and

causing testing of a subset of the plurality of templates for detecting a second target object, the testing comprising;

obtaining a second plurality of electronic images;

determining a region within each of the second plurality of electronic images, wherein the determined region corresponds to the region of interest of the first plurality of electronic images;

configuring a classifier for detecting the second target object based at least in part on the determined region; and

obtaining performance information related to the classifier in detecting the second target object from the second plurality of electronic images.

7. The computer-implemented method of claim 6, further comprising causing configuration of an other classifier for detecting a third target object, wherein the configuration of the other classifier is based at least in part on a template of the plurality of templates.

8. The computer-implemented method of claim 6, further comprising causing configuration of an other classifier for detecting a third target object, wherein the configuration of the classifier is based at least in part on the region of interest.

9. The computer-implemented method of claim 6, further comprising updating the plurality of templates based at least in part on the second plurality of electronic images.

10. The computer-implemented method of claim 6, wherein the generating of the plurality of templates comprises generating a sample from the region of interest and generating a template based on the sample.

11. A system comprising:

a data store configured to store computer-executable instructions; and

a hardware processor in communication with the data store, the hardware processor, configured to execute the computer-executable instructions to at least;

obtain a first set of images, wherein each image of the first set includes a first target object for detection;

identify a plurality of regions within each image of the first set, wherein a region of the plurality of regions includes at least a portion of the first target object;

for each region of the plurality of regions, determine an associated likelihood of the region showing at least part of the first target object based, at least in part, on a location of the region relative to a corresponding image in accordance with a location distribution of the first target object over the first set of images and on a measure of coverage by an unobstructed portion of the first target object relative to the area of the region;

generate a set of templates based at least in part on a subset of the plurality of regions, wherein the previously determined associated likelihood of each region of the subset exceeds a threshold value defining a high likelihood;

cause selection of a subset of the set of templates based at least on performance information related to detecting a second target object, wherein the second target object is detected from a second set of images using a first classifier, and wherein the first classifier is configured based at least in part on the selected subset of templates; and

cause a second classifier to be configured for detecting a third target object from a third set of images, wherein the second classifier is configured based at least in part on a template of the selected subset of templates.

12. The system of claim 11, wherein the first, second, and third target objects are a same target object.

13. The system of claim 11, wherein the region of the plurality of regions includes at least an obstructed portion of the first target object.

14. The system of claim 11, wherein identifying the plurality of regions comprises labeling an area within each image of the first set as containing at least part of the first target object.

15. The system of claim 14, wherein labeling the area comprises associating the area with a feature of the first target object.

16. The system of claim 11, wherein the threshold value is predetermined.

17. The system of claim 11, wherein the hardware processor is further configured to determine the associated likelihood of the region based on individual pixels of each image of the first set of images constituting a pixel of the first target object.

18. The system of claim 11, wherein generating the set of templates comprises generating a sample within a region of the identified plurality of regions and generating templates based on the sample.

19. The system of claim 11, wherein causing selection of the subset of the set of templates includes causing selection of the subset from all templates in the set of templates.

20. A non-transitory computer-readable storage medium storing computer-executable instructions that when executed by a processor perform operations comprising:

for each region of a plurality of regions within an image, determining an associated likelihood of the region showing at least part of a first target object based, at least in part, on a location of the region relative to the image in accordance with a location distribution of the first target object over a first plurality of images and on a measure of coverage by an identified unobstructed portion of the first target object relative to the area of the region, wherein the image belongs to the first plurality of images;

selecting a region of interest from the plurality of regions, wherein the associated likelihood of the region of interest exceeds a threshold value defining a high likelihood;

generating a plurality of templates based at least in part on the selected region of interest; and

causing testing of a subset of the plurality of templates for detecting a second target object, the testing comprising;

obtaining a second plurality of images;

determining a region within each of the second plurality of images, wherein the determined region corresponds to the region of interest of the first plurality of images;

configuring a classifier for detecting the second target object based at least in part on the determined region; and

obtaining performance information related to the classifier in detecting the second target object from the second plurality of images.

21. The non-transitory computer-readable storage medium of claim 20, wherein the operations further comprise updating the plurality of templates based at least in part on the second plurality of images.

22. The non-transitory computer-readable storage medium of claim 20, wherein the operations further comprise selecting the subset of templates from the plurality of templates based at least in part on the second plurality of images.

23. The non-transitory computer-readable storage medium of claim 20, wherein generating the plurality of templates comprises generating a sample from the region of interest and generating a template based on the sample.