SCENE ANALYSIS

• US 20070098222A1
• Filed: 10/24/2006
• Published: 05/03/2007
• Est. Priority Date: 10/31/2005
• Status: Abandoned Application

• Current Assignee: Sony United Kingdom Limited (Sony Group Corp.)
• Original Assignee: Sony United Kingdom Limited (Sony Group Corp.)
• Inventors: Haynes, Simon, Porter, Robert, Beresford, Ratna
• Application Number:
  US11/552,278
• Publication Number:
  US 20070098222A1
• Time in Patent Office: Days
• US Class Current: 382/103
• CPC Class Codes: G06V 20/53 Recognition of crowd images, e.g. recognition of crowd congestion
• 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 40/162 using pixel segmentation or colour matching

Abstract

Apparatus is arranged in operation to perform a method of estimating the number of individuals in a scene. The method comprises generating, for a plurality of image positions within at least a portion of a captured image of the scene, an edge correspondence value indicative of positional and angular correspondence with a representation of at least a partial outline of an individual. Analysis of the edge correspondence value is used to detect whether each of the plurality of image positions contributes to at least part of an image of an individual.

Claims

1. A method of estimating the number of individuals in an image, the method comprising the steps of:

(i) generating, for a plurality of image positions within at least a portion of a captured image of the scene, an edge correspondence value indicative of positional and angular correspondence with a template representation of at least a partial outline of an individual, and;

(ii) detecting whether image content at each of the image positions corresponds to at least a part of an image of an individual in response to said detected edge correspondence value.

2. A method according to claim 1, in which said step of generating the edge correspondence value comprises:

comparing, for an image position in said image, a plurality of edges derived from said captured image with at least a first edge angle template located with respect to that image position, said edge angle template relating expected edge angles to expected relative positions between said edges, the expected relative positions between said edges being representative of at least said partial outline of said individual.

3. A method according to claim 1, in which an edge angle template relating expected edge angles to expected relative positions between said edges comprises a spatial distribution of angular values over said edge angle template, such that said angular values are located with respect to positions representative of at least said partial outline of said individual where such corresponding angles are likely to be present.

4. A method according to claim 1, wherein said at least partial outline of said individual is an at least partial outline of a head.

5. A method according to claim 1, comprising the step of:

obtaining horizontal edge values and vertical edge values by respective application of a horizontal and a vertical spatial gradient operator to said portion of said captured image.

6. A method according to claim 1, comprising the step of:

further processing said horizontal edge values and vertical edge values in combination to generate edge magnitude values.

7. A method according to claim 1, comprising the step of:

obtaining edge angle estimates by analysis of corresponding vertical and horizontal edge values.

8. A method according to claim 7, comprising the step of:

obtaining edge angle estimates by applying an arctan function to a quotient of corresponding vertical and horizontal edge values.

9. A method according to claim 7, comprising the step of:

discarding edge angle estimates corresponding to low-magnitude edge values.

10. A method according to claim 7, comprising the step of:

evaluating edge angle estimates against an edge angle template as a function of the relative parallelism found between an edge angle estimate and said edge angle value located at a corresponding position on said template.

11. A method according to claim 10, comprising the steps of:

evaluating, within each of a plurality of zones of said edge angle template, said edge angle estimate most parallel to an edge angle value at the corresponding position on said edge angle template, and;

combining the differences in angular value between each such selected edge angle estimate and said corresponding edge angle template value for said plurality of zones to generate the edge correspondence value indicative of overall positional and angular correspondence with said edge angle template.

12. A method according to claim 7, comprising the step of:

quantising edge angle estimates and edge angle template values.

13. A method according to claim 1, in which said step of defining whether each of said plurality of image positions contributes to at least part of an image of said individual further comprises the step of satisfying one or more conditions selected from the list consisting of:

i. a body likelihood value exceeds a body value threshold;

ii. a head-centre likelihood value lies within the bounds of an upper and a lower head centre threshold;

iii. a head-top likelihood value exceeds a head-top value threshold;

iv. a head-sides likelihood value exceeds a head-sides value threshold; and

v. an edge mask convolution value exceeds an edge mask convolution value threshold.

14. A method according to claim 13, comprising the step of:

generating a body likelihood value for an image position in said scene by the summation of vertical edge values occurring in a region centred below that image position.

15. A method according to claim 13, comprising the step of:

generating a head-centre likelihood value for an image position in said scene by correlating edge magnitudes with a head-centre template positioned with respect to that image position, said head-centre template scoring positively in a central region of said head-centre template only.

16. A method according to claim 13, comprising the step of:

blurring horizontal edges and vertical edges to generate values adjacent to said edges, said values diminishing with distance from to said edges.

17. A method according to claim 16, comprising the step of:

generating a head-top likelihood value for an image position in said scene by correlating blurred horizontal edges with a head-top template positioned with respect to that image position, said template scoring positively in an upper region of said head-top template only, and negatively in a central region of said head-top template only.

18. A method according to claim 16, comprising the step of:

generating a head-sides likelihood value for a point in said scene by correlating blurred vertical edges with a head-sides template positioned with respect to that image position, said template scoring positively in side regions of said head-sides template only, and negatively in a central region of said head-sides template only.

19. A method according to claim 13, comprising the step of:

generating an edge mask convolution value for a point in said scene by convolving normalised horizontal and vertical edges with one or more respective horizontal and vertical edge masks, and selecting the largest output value as said edge mask convolution value.

20. A method according to claim 1, in which said captured image is first enhanced by the steps of:

generating a difference map between said captured image and a background image;

applying a low-pass filter to said background image to create a blurred background image; and

subtracting said blurred background image from said captured image, multiplying the result based upon said difference map values, and adding the output of said multiplication to said blurred background image.

21. A method according to claim 1, comprising the step of:

estimating the number of individuals in an image by counting those image positions, or localised groups of image positions, detected to be contributing to at least part of an image of an individual.

22. A method according to claim 21 comprising the step of:

estimating a change in said number of individuals in said image by comparing successive estimates of said number of individuals in respective successive images.

28. A data carrier comprising computer readable instructions that, when loaded into a computer, cause said computer to carry out the method of claim 1.

30. A data signal comprising computer readable instructions that, when received by a computer, cause said computer to carry out the method of claim 1.

32. Computer readable instructions that, when received by a computer, cause said computer to carry out the method of claim 1.

23. A data processing apparatus, arranged in operation to estimate said number of individuals in a scene, said apparatus comprising;

an analyser operable to generate, for a plurality of image positions within at least a portion of a captured image of said scene, an edge correspondence value indicative of positional and angular correspondence with a template representation of at least a partial outline of an individual, and

logic operable to detect whether image content at each of said image positions corresponds to at least a part of an image of an individual in response to said detected edge correspondence value.

24. A data processing apparatus according to claim 23, comprising an edge angle matcher arranged in operation to compare a plurality of edges derived from said image data with at least a first edge angle template located with respect to that image position, said edge angle template relating expected edge angles to expected relative positions between said edges, said expected relative positions between said edges being representative of at least said partial outline of said individual, and said edge angle matcher outputting said edge correspondence value based upon said comparison.

25. A data processing apparatus according to claim 23, further comprising an edge angle calculator operable to apply an arctan function to a quotient of corresponding horizontal and vertical edge values.

26. A data processing apparatus according to claim 23, in which said edge angle matcher is arranged in operation to evaluate edge angle estimates against an edge angle template as a function of the relative parallelism found between an edge angle estimate and said edge angle value located at a corresponding position on said template.

27. A data processing apparatus according to claim 23, in which said edge angle matcher is arranged in operation to select, within a plurality of zones of said edge angle template, said edge angle estimate evaluated as most parallel to an edge angle value at the corresponding position on said edge angle template, and combine the differences between the most parallel edge angle estimate and said edge angle value at the corresponding position for said plurality of zones to generate said edge correspondence value, being indicative of overall positional and angular correspondence with said edge angle template.

29. A data carrier comprising computer readable instructions that, when loaded into a computer, cause said computer to operate as a data processing apparatus according to claim 23.

31. A data signal comprising computer readable instructions that, when received by a computer, cause said computer to operate as a data processing apparatus according to claim 23.

33. Computer readable instructions that, when received by a computer, cause said computer to operate as a data processing apparatus according to claim 23.