Detection method and detection device

• US 11,210,513 B2
• Filed: 10/22/2019
• Issued: 12/28/2021
• Est. Priority Date: 10/30/2018
• Status: Active Grant

• Current Assignee: Fujitsu Limited
• Original Assignee: Fujitsu Limited
• Inventors: Endoh, Toshio, Kato, Keizo
• Primary Examiner(s): Lee, John W
• Application Number:
  US16/660,743
• Publication Number:
  US 20200134313A1
• Time in Patent Office: 798 Days
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06F 18/2413 based on distances to training or reference patterns
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06N 3/084 Backpropagation, e.g. using gradient descent
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06V 10/25 Determination of region of interest [ROI] or a volume of interest [VOI]
• CPC Class Codes: G06V 10/421 by analysing segments intersecting the pattern
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 10/774 Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06V 20/00 Scenes; Scene-specific elements control of digital cameras H04N23/60
• CPC Class Codes: G06V 40/103 Static body considered as a whole, e.g. static pedestrian or occupant recognition

Abstract

A computer-implemented detection method includes, in response to inputting a first image including a region of one or more objects to a learned model, identifying a first entire image corresponding to entirety of a first object as a detection candidate, the learned model being generated by learning training data including an image corresponding to a part of an object and an entire image corresponding to entirety of the object, detecting an existing region of the first target object in the first image in accordance with a comparison between the identified first entire image and the region of the one or more target objects, and determining, based on a specific image obtained by invalidating the existing region in the first image, whether another target object is included in the first image.

Claims

1. A computer-implemented detection method comprising:

in response to inputting a first image to a learned model, identifying a first estimated image from the input first image by using the learned model, the first image being an image including a region of a plurality of target objects, the first estimated image being an estimated image corresponding to entirety of a first target object being one of the plurality of target objects, the learned model being a model generated by learning training data including a partial image indicating a part of an object and a corresponding entire image indicating entirety of the object;

detecting a first region of the first target object in the first image in accordance with a comparison between the identified first estimated image and the first image, the first region being a region including a contour of entirety of the first target object;

identifying, based on a second image obtained by invalidating the first region from among the first image, a second estimated image with respect to a second target object by using the learned model, the second estimated image being an estimated image corresponding to entirety of the second target object, the second target object being one of the plurality of target objects other than the first target object; and

detecting a second region of the second target object in accordance with a comparison between the identified second estimated image and the second image, the second region being a region including a contour of entirety of the second target object, the comparison between the identified second estimated image and the second image being configured to exclude a region corresponding to the invalidated first region from the comparison between the identified second estimated image and the second image.

2. The detection method according to claim 1, wherein

the identifying of the first entire image includes identifying a plurality of entire images corresponding to respective entirety of a plurality of target objects including the first target object, and

the detecting of the existing region includes selecting the first target object based on a difference between each of the plurality of entire images and the region of the one or more target objects.

3. The detection method according to claim 2, wherein

the determining includes detecting a second target object among the plurality of target objects based on a difference between each of the plurality of entire images other than the first entire image and the region of the one or more target objects other than the existing region.

4. The detection method according to claim 1, wherein

the detecting of the existing region includes determining the existing region based on a size of a difference between the first entire image and the region of the one or more target objects.

5. A detection device comprising:

a memory; and

a processor coupled to the memory and the processor configured to;

in response to inputting a first image to a learned model, perform identification of a first estimated image from the input first image by using the learned model, the first image being an image including a region of a plurality of target objects, the first estimated image being an estimated image corresponding to entirety of a first target object being one of the plurality of target objects, the learned model being a model generated by learning training data including a partial image indicating a part of an object and a corresponding entire image indicating entirety of the object,

perform detection of a first region of the first target object in the first image in accordance with a comparison between the identified first estimated image and the first image, the first region being a region including a contour of entirety of the first target object,

identify, based on a second image obtained by invalidating the first region from among the first image, a second estimated image with respect to a second target object by using the learned model, the second estimated image being an estimated image corresponding to entirety of the second target object, the second target object being one of the plurality of target objects other than the first target object, and

detect a second region of the second target object in accordance with a comparison between the identified second estimated image and the second image, the second region being a region including a contour of entirety of the second target object, the comparison between the identified second estimated image and the second image being configured to exclude a region corresponding to the invalidated first region from the comparison between the identified second estimated image and the second image.

6. The detection device according to claim 5, wherein

the identification of the first entire image includes identifying a plurality of entire images corresponding to respective entirety of a plurality of target objects including the first target object, and

the detection of the existing region includes selecting the first target object based on a difference between each of the plurality of entire images and the region of the one or more target objects.

7. The detection device according to claim 6, wherein

the determination includes detecting a second target object among the plurality of target objects based on a difference between each of the plurality of entire images other than the first entire image and the region of the one or more target objects other than the existing region.

8. The detection device according to claim 5, wherein

the detection of the existing region includes determining the existing region based on a size of a difference between the first entire image and the region of the one or more target objects.

9. A non-transitory computer-readable medium storing instructions executable by one or more computers, the instructions comprising:

one or more instructions for identifying, in response to inputting a first image to a learned model, a first estimated image from the input first image by using the learned model, the first image being an image including a region of a plurality of target objects, the first estimated image being an estimated image corresponding to entirety of a first target object being one of the plurality of target objects, the learned model being a model generated by learning training data including a partial image indicating a part of an object and a corresponding entire image indicating entirety of the object;

one or more instructions for detecting a first region of the first target object in the first image in accordance with a comparison between the identified first estimated image and the first image, the first region being a region including a contour of entirety of the first target object;

one or more instructions for identifying, based on a second image obtained by invalidating the first region from among the first image, a second estimated image with respect to a second target object by using the learned model, the second estimated image being an estimated image corresponding to entirety of the second target object, the second target object being one of the plurality of target objects other than the first target object; and

detecting a second region of the second target object in accordance with a comparison between the identified second estimated image and the second image, the second region being a region including a contour of entirety of the second target object, the comparison between the identified second estimated image and the second image being configured to exclude a region corresponding to the invalidated first region from the comparison between the identified second estimated image and the second image.

10. The medium according to claim 9, wherein

the identifying of the first entire image includes identifying a plurality of entire images corresponding to respective entirety of a plurality of target objects including the first target object, and

the detecting of the existing region includes selecting the first target object based on a difference between each of the plurality of entire images and the region of the one or more target objects.

11. The medium according to claim 10, wherein

the determining includes detecting a second target object among the plurality of target objects based on a difference between each of the plurality of entire images other than the first entire image and the region of the one or more target objects other than the existing region.

12. The medium according to claim 9, wherein

the detecting of the existing region includes determining the existing region based on a size of a difference between the first entire image and the region of the one or more target objects.