OBJECT DETECTION APPARATUS

• US 20110026812A1
• Filed: 04/06/2009
• Published: 02/03/2011
• Est. Priority Date: 04/09/2008
• Status: Active Grant

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Ikeda, Hiroo
• Granted Patent: US 8,311,281 B2
• Time in Patent Office: Days
• US Class Current: 382/159
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/143 involving probabilistic approaches, e.g. Markov random field [MRF] modelling
• CPC Class Codes: G06V 10/255 Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
• CPC Class Codes: G06V 10/28 Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns

Abstract

In order to improve object detection precision, an object detection apparatus includes a posterior probability calculation portion utilizing an occurrence probability of a background and a foreground acquired by utilizing a characteristic quantity extracted from each pixel of an input image and a probability density function, a posterior probability of the previous background and foreground, and a conditional probability indicating a relation of an event (background or foreground) to the vicinity of an attentive pixel in a spatial direction and a relation of an event to the vicinity of the attentive pixel in a temporal direction so as to calculate a posterior probability of the background and the foreground from a probability model utilizing a tendency that the same event appears together in the vicinity of the attentive pixel in the spatial and temporal directions; and an object determination portion for determining an object from comparison between the posterior probabilities of the background and the foreground.

Claims

1. An object detection apparatus comprising:

a posterior probability calculation portion utilizing an occurrence probability of a background and a foreground acquired by utilizing a characteristic quantity extracted from each pixel of an input image and a probability density function, a posterior probability of the background and the foreground for the previous input image to the input image, and a conditional probability indicating a relation of an event (background or foreground) to the vicinity of an attentive pixel in a spatial direction and a relation of an event to the vicinity of the attentive pixel in a temporal direction so as to calculate a posterior probability of the background and the foreground from a probability model set up based on a tendency that the same event appears in the vicinity of the attentive pixel in the spatial and temporal directions; and

an object determination portion for determining an object from comparison of the posterior probability of the background with that of the foreground.

2. The object detection apparatus according to claim 1, wherein the conditional probability is an occurrence probability of an event of a peripheral pixel in the vicinity of the spatial direction when an event of the attentive pixel (background or foreground) occurs, and an occurrence probability of an event of the present attentive pixel in the vicinity of the temporal direction when an event of the attentive pixel of the previous input image occurs.

3. The object detection apparatus according to claim 2, wherein in the vicinity of the spatial direction, the conditional probability is set higher in the case that the event of the attentive pixel, is the same as the event of the peripheral pixel than in the case that they are different;

and in the vicinity of the temporal direction, the conditional probability is set higher in the case that the event of the attentive pixel of the previous input image is the same as the event of the present attentive pixel than in the case that they are different.

4. The object detection apparatus according to claim 2, wherein from a determination result of the object determination portion, the conditional probability sequentially counts a number with respect to totally four cases that the peripheral pixel is determined as the background or the foreground when the attentive pixel is determined as the background or the foreground, and other totally four cases that the present attentive pixel is determined as the background or the foreground when the attentive pixel of the previous input image is determined as the background or the foreground, and utilizes the number to self-update.

5. The object detection apparatus according to claim 4, wherein the conditional probability should neither be lower than ½

when the event of the attentive pixel is the same as the event of the peripheral pixel in the vicinity of the spatial direction nor be lower than ½

when the event of the attentive pixel of the previous input image is the same as the event of the present attentive pixel in the vicinity of the temporal direction.

6. The object detection apparatus according to claim 1, wherein the probability density function of the background in the probability density functions observes a determination result of the object determination portion in a time-series manner, and is estimated from the characteristic quantity sequentially accumulated as the background has changed in trend if an appearance frequency of the foreground is high, while from the characteristic quantity accumulated when the determination result is determined as the background by the object determination portion as the background has not changed in trend if the appearance frequency of the foreground is low.

7. The object detection apparatus according to claim 1, wherein the characteristic quantity is a two-dimensional or multidimensional characteristic vector combining a plurality of one-dimensional characteristic quantities.

8. The object detection apparatus according to claim 1, wherein the characteristic quantity is a characteristic quantity and a characteristic vector with a block acquired from a local area set up in the image as a unit.

9. The object detection apparatus according to claim 1, wherein the probability density function of the foreground in the probability density functions is uniformly distributed.

10. The object detection apparatus according to claim 1, wherein the probability density function of the foreground in the probability density functions is a probability density function estimated by analyzing the target object to be detected in advance.

11. The object detection apparatus according to claim 1, wherein the object determination portion compares the posterior probabilities of the background and the foreground acquired by the posterior probability calculation portion to determine the object as the background if the posterior probability of the background is higher and as the foreground if the posterior probability of the foreground is higher.

12. An object detection apparatus comprising:

a posterior probability calculation portion for calculating a posterior probability of a foreground for an attentive pixel and a posterior probability of a background for the attentive pixel as the posterior probability of the foreground and the posterior probability of the background for the attentive pixel of a present image included in a moving image; and

an object determination portion for determining an object based on comparison of the posterior probability of the foreground for the attentive pixel with the posterior probability of the background for the attentive pixel,

the posterior probability calculation portion calculating the posterior probability of the background for the attentive pixel based on;

an occurrence probability of the background for the attentive pixel as the occurrence probability of the background of the attentive pixel;

an occurrence probability of the background for a peripheral pixel as the occurrence probability of the background of the peripheral pixel which is a pixel of the present image and located in the vicinity of the attentive pixel;

a first spatial relation probability indicating a probability that the peripheral pixel becomes the background when the attentive pixel is the background;

an occurrence probability of the foreground for the peripheral pixel as the occurrence probability of the foreground of the peripheral pixel;

a second spatial relation probability indicating a probability that the peripheral pixel becomes the foreground when the attentive pixel is the background;

a posterior probability of the background for a past pixel as the posterior probability of the background of the past pixel which is a pixel of a past image included in the moving image and corresponds to the attentive pixel;

a first temporal relation probability indicating a probability that the attentive pixel becomes the background when the past pixel is the background;

a posterior probability of the foreground for the past pixel as the posterior probability of the foreground of the past pixel; and

a second temporal relation probability indicating a probability that the attentive pixel becomes the background when the past pixel is the foreground,

the posterior probability calculation portion calculating the posterior probability of the foreground for the attentive pixel based on;

an occurrence probability of the foreground for the attentive pixel as the occurrence probability of the foreground of the attentive pixel;

the occurrence probability of the background for the peripheral pixel;

a third spatial relation probability indicating a probability that the peripheral pixel becomes the background when the attentive pixel is the foreground;

the occurrence probability of the foreground for the peripheral pixel;

a fourth spatial relation probability indicating a probability that the peripheral pixel becomes the foreground when the attentive pixel is the foreground;

the posterior probability of the background for the past pixel;

a third temporal relation probability indicating a probability that the attentive pixel becomes the foreground when the past pixel is the background;

the posterior probability of the foreground for the past pixel; and

a fourth temporal relation probability indicating a probability that the attentive pixel becomes the foreground when the past pixel is the foreground,

the first spatial relation probability being higher than the second spatial relation probability;

the fourth spatial relation probability being higher than the third spatial relation probability;

the first temporal relation probability being higher than the third temporal relation probability; and

the fourth temporal relation probability being higher than the second temporal relation probability.

13. The object detection apparatus according to claim 12 further comprising a relation probability calculation portion,

wherein the sum of the first spatial relation probability and the second spatial relation probability is 1, the sum of the fourth spatial relation probability and the third spatial relation probability is 1, the sum of the first temporal relation probability and the third temporal relation probability is 1, and the sum of the fourth temporal relation probability and the second temporal relation probability is 1;

and

wherein the relation probability calculation portion calculates a first spatial relation probability value, a fourth spatial relation probability value, a first temporal relation probability value, and a fourth temporal relation probability value;

sets the first spatial relation probability at the first spatial relation probability value when the first spatial relation probability value is equal to or higher than ½

, sets the first spatial relation probability at a value equal to or higher than ½

when the first spatial relation probability value is lower than ½

, sets the fourth spatial relation probability at the fourth spatial relation probability value when the fourth spatial relation probability value is equal to or higher than ½

, sets the fourth spatial relation probability at a value equal to or higher than ½

when the fourth spatial relation probability value is lower than ½

, sets the first temporal relation probability at the first temporal relation probability value when the first temporal relation probability value is equal to or higher than ½

, sets the first temporal relation probability at a value equal to or higher than ½

when the first temporal relation probability value is lower than ½

, sets the fourth temporal relation probability at the fourth temporal relation probability value when the fourth temporal relation probability value is equal to or higher than ½

, and sets the fourth temporal relation probability at a value equal to or higher than ½

when the fourth temporal relation probability value is lower than ½

.

14. The object detection apparatus according to claim 12 further comprising an occurrence probability acquisition portion, wherein the occurrence probability acquisition portion acquires the occurrence probability of the background for the attentive pixel and the occurrence probability of the foreground for the attentive pixel based on the characteristic quantity of the attentive pixel, and acquires the occurrence probability of the background for the peripheral pixel and the occurrence probability of the foreground for the peripheral pixel based on the characteristic quantity of the peripheral pixel.

15. A computer program product comprising computer implementable instructions to make a computer function as:

a posterior probability calculation portion utilizing an occurrence probability of a background and a foreground acquired by utilizing a characteristic quantity extracted from each pixel of an input image and a probability density function, a posterior probability of the background and the foreground for the previous input image to the input image, and a conditional probability indicating a relation of an event (background or foreground) to the vicinity of an attentive pixel in a spatial direction and a relation of an event to the vicinity of the attentive pixel in a temporal direction so as to calculate a posterior probability of the background and the foreground from a probability model set up based on a tendency that the same event appears in the vicinity of the attentive pixel in the spatial and temporal directions; and

an object determination portion for determining an object from comparison of the posterior probability of the background with that of the foreground.

16. The computer program product according to claim 15, wherein the conditional probability is an occurrence probability of an event of a peripheral pixel in the vicinity of the spatial direction when an event of the attentive pixel (background or foreground) occurs, and an occurrence probability of an event of the present attentive pixel in the vicinity of the temporal direction when an event of the attentive pixel of the previous input image occurs.

17. An object detection method comprising the steps of:

calculating a posterior probability by utilizing an occurrence probability of a background and a foreground acquired by utilizing a characteristic quantity extracted from each pixel of an input image and a probability density function, a posterior probability of the background and the foreground for the previous input image to the input image, and a conditional probability indicating a relation of an event (background or foreground) to the vicinity of an attentive pixel in a spatial direction and a relation of an event to the vicinity of the attentive pixel in a temporal direction so as to calculate the posterior probability of the background and the foreground from a probability model set up based on a tendency that the same event appears in the vicinity of the attentive pixel in the spatial and temporal directions; and

determining an object from comparison of the posterior probability of the background with that of the foreground.

18. The object detection method according to claim 17, wherein the conditional probability is an occurrence probability of an event of a peripheral pixel in the vicinity of the spatial direction when an event of the attentive pixel (background or foreground) occurs, and an occurrence probability of an event of the present attentive pixel in the vicinity of the temporal direction when an event of the attentive pixel of the previous input image occurs.

19. The object detection method according to claim 18, wherein in the vicinity of the spatial direction, the conditional probability is set higher in the case that the event of the attentive pixel is the same as the event of the peripheral pixel than in the case that they are different;

and in the vicinity of the temporal direction, the conditional probability is set higher in the case that the event of the attentive pixel of the previous input image is the same as the event of the present attentive pixel than in the case that they are different.

20. The object detection method according to claim 18, wherein from a result by the determining, the conditional probability sequentially counts a number with respect to totally four cases that the peripheral pixel is determined as the background or the foreground when the attentive pixel is determined as the background or the foreground, and other totally four cases that the present attentive pixel is determined as the background or the foreground when the attentive pixel of the previous input image is determined as the background or the foreground, and utilizes the number to self-update.

21. The object detection method according to claim 20, wherein the conditional probability should neither be lower than ½

when the event of the attentive pixel is the same as the event of the peripheral pixel in the vicinity of the spatial direction nor be lower than ½

when the event of the attentive pixel of the previous input image is the same as the event of the present attentive pixel in the vicinity of the temporal direction.

22. The object detection method according to claim 17, wherein the probability density function of the background in the probability density functions observes a result by the determining in a time-series manner, and is estimated from the characteristic quantity sequentially accumulated as the background has changed in trend if an appearance frequency of the foreground is high, while from the characteristic quantity accumulated when the determination result is determined as the background by determining as the background has not changed in trend if the appearance frequency of the foreground is low.

23. The object detection method according to claim 17, wherein the characteristic quantity is a two-dimensional or multidimensional characteristic vector combining a plurality of one-dimensional characteristic quantities.

24. The object detection method according to claim 17, wherein the characteristic quantity is a characteristic quantity and a characteristic vector with a block acquired from a local area set up in the image as a unit.

25. The object detection method according to claim 17, wherein the probability density function of the foreground in the probability density functions is uniformly distributed.

26. The object detection method according to claim 17, wherein the probability density function of the foreground in the probability density functions is a probability density function estimated by analyzing the target object to be detected in advance.

27. The object detection method according to claim 17, wherein in the determining, comparison is made between the posterior probabilities of the background and the foreground acquired in the calculating of the posterior probability to determine the object as the background if the posterior probability of the background is higher and as the foreground if the posterior probability of the foreground is higher.

28. An object detection apparatus comprising:

a posterior probability calculation means for utilizing an occurrence probability of a background and a foreground acquired by utilizing a characteristic quantity extracted from each pixel of an input image and a probability density function, a posterior probability of the background and the foreground for the previous input is image to the input image, and a conditional probability indicating a relation of an event (background or foreground) to the vicinity of an attentive pixel in a spatial direction and a relation of an event to the vicinity of the attentive pixel in a temporal direction so as to calculate a posterior probability of the background and the foreground from a probability model set up based on a tendency that the same event appears in the vicinity of the attentive pixel in the spatial and temporal directions; and

an object determination means for determining an object from comparison of the posterior probability of the background with that of the foreground.