Human detection in high density crowds

• US 10,402,633 B2
• Filed: 05/23/2016
• Issued: 09/03/2019
• Est. Priority Date: 05/23/2016
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Tang, Shaopeng, Chen, Yurong
• Primary Examiner(s): Safaipour, Bobbak
• Application Number:
  US15/521,512
• Publication Number:
  US 20180189557A1
• Time in Patent Office: 1,198 Days
• CPC Class Codes: G06T 2207/10012 Stereo images
• CPC Class Codes: G06T 2207/20021 Dividing image into blocks, subimages or windows
• CPC Class Codes: G06T 2207/30196 Human being; Person
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06V 40/10 Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
• CPC Class Codes: G06V 40/20 Movements or behaviour, e.g. gesture recognition recognition of facial expressions G06V40/16
• CPC Class Codes: H04N 13/204 using stereoscopic image cameras stereoscopic photography G03B35/00
• CPC Class Codes: H04N 7/18 Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast

Abstract

The present disclosure describes a non-learning based process and apparatus for detecting humans in an image. This may include receiving an image that has pixel distance information from a camera and using that to determine a height of the pixel above a ground surface. One or more regions may then be identified that may include a head and shoulders of an individual in the image. A multiple threshold technique may be used to remove some background regions, and a mean-shift technique used to find the local highest regions that may be combination of head and shoulders of the person. In embodiments, the view angle and/or the height of the camera may not be fixed.

Claims

1. An apparatus to detect a person, the apparatus comprising:

one or more computer processors;

and

a region proposal module, communicatively coupled to the one or more processors, to identify within an image one or more regions that may depict a person;

wherein the region proposal module is to;

divide the image into a plurality of cells in a uniform grid;

calculate an average height value of each cell;

apply a mean-shift operation to each cell;

identify one or more candidate regions based at least upon one or more cluster centers identified by the mean-shift operation, wherein the cluster centers represent local highest regions in the image;

validate the one or more candidate regions; and

output one or more indications of the identified one or more regions.

2. The apparatus of claim 1, wherein each of a subset of the plurality of cells comprises a number of pixels greater than a first threshold value having respective corresponding height values above a second threshold value, wherein the each of the subset further defines one or more candidate regions.

3. The apparatus of claim 1, wherein to validate, the region proposal module is to identify, based at least on a template matching or a learning-based system, whether the one or more candidate regions represents a person.

4. The apparatus of claim 3, wherein represents a person comprises represents a head and shoulders combination of a person.

5. The apparatus of claim 1, wherein the region proposal module is to determine a height value for each of objects of a plurality of pixels of the image using a predetermined map associated with a camera.

6. The apparatus of claim 5, wherein the predetermined map is based at least upon a view angle of the camera.

7. The apparatus of claim 6, wherein the camera comprises a 3-D camera.

8. The apparatus of claim 1, wherein a camera is set to a view angle selected from a plurality of settable view angles.

9. The apparatus of claim 1, further comprising a height conversion module, communicatively coupled to the one or more processors, to process an image captured by a camera and to provide a height value for each object associated, respectively, by a plurality of pixels of the image, relative to a ground surface;

wherein the height conversion module is to extract, from the image, distance values of the objects of the pixels to the camera, and respectively convert the distance values associated with the plurality of pixels of the image to corresponding height values of the objects of the plurality of the pixels of the received image, relative to a ground surface.

10. A computer implemented method to detect a person, comprising:

receiving, by a computing device, an image captured by a camera, wherein the image includes a distance value for each of pixels of the image, respectively, associated with a plurality of objects to the camera;

extracting, by the computing device, distance values of the objects of the pixels to the camera;

respectively converting, by the computing device, the distance values associated with the plurality of objects of the pixels of the image to corresponding height values for the plurality of the objects of the pixels of the image relative to a ground surface;

analyzing, by the computing device, the height values of the objects of the pixels relative to the ground surface;

dividing, by the computing device, the image into a plurality of cells in a uniform grid;

calculating, by the computing device, an average height value of each cell;

applying, by the computing device, a mean-shift operation to each cell;

identifying, by the computing device, one or more regions within the image that may depict a person; and

outputting, by the computing device, one or more indications of the identified one or more regions.

11. One or more non-transitory computer-readable media comprising instructions that cause a computing device, in response to execution of the instructions by the computing device, to:

receive, by a computing device, an image captured by a camera;

divide, by the computing device, the image into a plurality of cells in a uniform grid;

calculate, by the computing device, an average height value of each cell;

apply, by the computing device, a mean-shift operation to each cell;

identify one or more candidate regions based at least upon one or more cluster centers identified by the mean-shift operation, wherein the cluster centers represent local highest regions in the image;

identify one or more regions based at least upon a validated one or more candidate regions; and

output, by the computing device, one or more indications of the identified one or more regions.

12. The one or more non-transitory computer-readable media of claim 11, wherein each of a subset of the plurality of cells comprises a number of pixels greater than a first threshold value having respective corresponding height values above a second threshold value, wherein the each of the subset further defines one or more candidate regions.

13. The one or more non-transitory computer-readable media of claim 11, wherein to validate the one or more candidate regions further comprises to identify, by the computing device, based at least on template matching, whether the one or more candidate regions is representing a person.

14. The one or more non-transitory computer-readable media of claim 13, wherein representing a person comprises representing a head and shoulders combination of a person.

15. The one or more non-transitory computer-readable media of claim 11, wherein to respectively convert the distance values to corresponding height values further comprises to determine a height value for each object associated with a plurality of pixels of the image using a conversion map associated with the camera.

16. The one or more non-transitory computer-readable media of claim 15, wherein the conversion map is based at least upon a view angle of the camera.

17. The one or more non-transitory computer-readable media of claim 11, wherein the image includes a distance value for each of a plurality of objects associated, respectively, with pixels of the image to the camera, and the computing device is further caused to extract distance values of the plurality of objects associated, respectively, with the pixels to the camera;

and respectively convert the distance values associated with the plurality of objects of the pixels of the image to corresponding height values for the plurality of the objects of the pixels of the image relative to a ground surface.

18. An apparatus to detect a person, the apparatus comprising:

means for receiving an image captured by a camera, wherein the image includes a distance value for a plurality of objects associated, respectively, with pixels of the image to the camera;

means for extracting distance values of the objects of the pixels to the camera;

means for respectively converting the distance values associated with the plurality of objects of the pixels of the image to corresponding height values for the plurality of the objects of the pixels of the image relative to a ground surface;

means for dividing the image into a plurality of cells in a uniform grid;

means for calculating an average height value of each cell;

means for applying a mean-shift operation to each cell;

means for identifying one or more candidate regions based at least upon one or more cluster centers identified by the mean-shift operation, wherein the cluster centers represent local highest regions in the image;

means for validating the one or more candidate regions;

means for identifying one or more regions based at least upon the validated one or more candidate regions;

means for analyzing the height values of the objects of the pixels relative to the ground surface;

means for identifying, based on the analysis, one or more regions within the image that may depict a person; and

means for outputting one or more indications of the identified one or more regions.

19. The apparatus of claim 18, wherein each of a subset of the plurality of cells comprises a number of pixels greater than a first threshold value having respective corresponding height values above a second threshold value, wherein the each of the subset further defines one or more candidate regions.

20. The apparatus of claim 19, wherein means for validating the one or more candidate regions further comprises means for identifying, based at least on template matching, whether the one or more candidate regions is representing a person.