Object detection method, object detection apparatus, and program

• US 10,395,103 B2
• Filed: 12/01/2015
• Issued: 08/27/2019
• Est. Priority Date: 03/30/2012
• Status: Active Grant

• Current Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Original Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Inventors: Tsukamoto, Kenji, Torii, Kan, Matsugu, Masakazu
• Primary Examiner(s): Rudolph, Vincent
• Assistant Examiner(s): Patel, Pinalben
• Application Number:
  US14/956,134
• Publication Number:
  US 20160086024A1
• Time in Patent Office: 1,365 Days
• CPC Class Codes: G06T 3/40 Scaling of whole images or parts thereof, e.g. expanding or contracting
• CPC Class Codes: G06V 10/255 Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
• CPC Class Codes: G06V 30/2504 Coarse or fine approaches, e.g. resolution of ambiguities or multiscale approaches
• CPC Class Codes: G06V 40/10 Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands

Abstract

An object detection method includes an image acquisition step of acquiring an image including a target object, a layer image generation step of generating a plurality of layer images by one or both of enlarging and reducing the image at a plurality of different scales, a first detection step of detecting a region of at least a part of the target object as a first detected region from each of the layer images, a selection step of selecting at least one of the layer images based on the detected first detected region and learning data learned in advance, a second detection step of detecting a region of at least a part of the target object in the selected layer image as a second detected region, and an integration step of integrating a detection result detected in the first detection step and a detection result detected in the second detection step.

Claims

1. An object detection method comprising:

acquiring an image;

generating a plurality of layer images by enlarging or reducing the image;

detecting a first object from at least one of the plurality of layer images;

estimating a specific object based on the detected first object;

selecting at least one of the plurality of layer images based on the estimated specific object;

detecting a second object larger than the detected first object in the selected layer image;

estimating the specific object based on the detected second object; and

determining the specific object based on the specific object estimated based on the first object and the specific object estimated based the second object.

2. The object detection method according to claim 1, wherein the estimating the specific object with use of the first object includes estimating one specific object;

the estimating the specific object with use of the second object includes estimating a plurality of the specific objects; and

a specific object having a highest overlap ratio with the one specific object is determined, from among the plurality of the specific objects estimated with use of the second object, as the specific object.

3. The object detection method according to claim 1, wherein the at least one of the layer images is selected based on both of a width and a height of the specific object estimated with use of the first object.

4. The object detection method according to claim 1, wherein the at least one of the layer images is selected based on a matrix corresponding to a size of the specific object estimated with use of the first object and a coefficient learned in advance by a learning data.

5. The abject detection method according to claim 1, wherein the specific object is an object in which a positional relation between the first object and the second object is specified.

6. The object detection method according to claim 5, wherein the specific object is a head portion region of a person, the first object is a head portion surrounding region of a person, and the second object is an entire body of a person.

7. The object detection method according to claim 1 further comprising generating a feature amount from each of the plurality of layer images, wherein the first object is detected based on the generated feature amount.

8. The object detection method according to claim 1, wherein the second object is detected by sliding a window of a model which is learned in advance over the selected layer image.

9. The object detection method according to claim 1, wherein the first object and the second object are each detected with use of respectively different models.

10. The object detection method according to claim 1 further comprising outputting the determined specific region.

11. The object detection method according to claim 1, wherein the generating the plurality of layer images includes generating a feature amount for each of the layer images.

12. The object detection method according to claim 1, wherein the determining the specific object includes integrating the specific object estimated based on the first object and the specific object estimated based the second object.

14. A non-transitory computer readable storage medium storing a program that causes a computer to perform the object detection method according to claim 1.

13. An object detection apparatus comprising:

a memory; and

a processor in communication with the memory, the processor functions as;

an acquisition unit configured to acquire an image;

a generation unit configured to generate a plurality of layer images by enlarging or reducing the image;

a first detection unit configured to detect a first object from at least one of the plurality of layer images;

a first estimation unit configured to estimate a specific object based on the detected first object;

a layer limitation unit configured to limit at least one of the plurality of layer images on which detection is to be performed based on the estimated specific object;

a second detection unit configured to detect a second object larger than the first object in the limited layer image;

a second estimation unit configured to estimate the specific object based on the detected second object; and

a determination unit configured to determine the specific object based on the specific object estimated by the first estimation unit and the specific object estimated by the second estimation unit.