Apparatus for image recognition

• US 20090279738A1
• Filed: 05/07/2009
• Published: 11/12/2009
• Est. Priority Date: 05/08/2008
• Status: Active Grant

• Current Assignee: DENSO Corporation
• Original Assignee: DENSO Corporation
• Inventors: Sasaki, Mikio
• Granted Patent: US 8,238,606 B2
• Time in Patent Office: Days
• US Class Current: 382/103
• CPC Class Codes: G06V 20/58 Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
• CPC Class Codes: G06V 40/103 Static body considered as a whole, e.g. static pedestrian or occupant recognition
• CPC Class Codes: H04N 19/51 Motion estimation or motion compensation

Abstract

An image recognition apparatus includes an image recognition unit, an evaluation value calculation unit, and a motion extraction unit. The image recognition unit uses motion vectors that are generated in the course of coding image data into MPEG format data or in the course of decoding the MPEG coded data by the evaluation value calculation unit and the motion extraction unit as well as two dimensional DCT coefficients and encode information such as picture types and block types for generating the evaluation values that represent feature of the image. The apparatus further includes an update unit for recognizing the object in the image based on the determination rules for a unit of macro block. The apparatus can thus accurately detect the motion of the object based on the evaluation values derived from DCT coefficients even when generation of the motion vectors is difficult.

Claims

1. An image recognition apparatus for recognizing an object by using encode information at one of image encoding and image decoding based on a motion-compensating prediction and a discrete cosine transform, the apparatus comprising:

an evaluation value calculation unit for calculating multiple kinds of evaluation values based on the encodes information; and

a determination unit for, according to a determination rule that defines relationships between (a) multiple objects that are to be extracted from an, input image and (b) the multiple evaluation values that are calculated by the evaluation value calculation unit, determining, for each of the multiple pixel blocks, which one of multiple pixel blocks corresponds to which one of the multiple objects, wherein

the evaluation value calculation unit at least uses, as the encode information, a two dimensional DCT coefficient derived from the discrete cosine transform.

2. The image recognition apparatus of claim 1 further comprising:

a first evaluation unit in the evaluation value calculation unit for calculating, as one of the multiple evaluation values, a basic evaluation value based on the two dimensional DCT coefficient for each of the multiple pixel blocks that is used as a unit of coding.

3. The image recognition apparatus of claim 2, wherein

the first evaluation unit calculates, as one of the multiple basic evaluation value, a sum total of alternate power components in the two dimensional DCT coefficient.

4. The image recognition apparatus of claim 3, wherein

the first evaluation unit calculates, as one of the multiple basic evaluation value, a sum total, of low frequency power components in the two dimensional DCT coefficient.

5. The image recognition apparatus of claim 2 further comprising:

a second evaluation unit in the evaluation value calculation unit for calculating, as one of the evaluation values, a compound evaluation value by using an evaluation function that defines a combination method of the basic evaluation values that are calculated by the first evaluation unit for each of specified object areas.

6. The image recognition apparatus of claim 5 further comprising:

an evaluation setting unit for setting the object area, the evaluation function and the determination rule to be applied to the object area based on a determination result of the determination unit

7. The image recognition apparatus of claim 6, wherein

the evaluation setting unit sets, based on models of road structures and scenery for determining a scene in the input image, the object area as a high possibility area that has a higher possibility of appearance of specified type objects.

8. The image recognition apparatus of claim 6, wherein

the evaluation setting unit sets the object area according to the determination result that includes positions and behaviors of the specified type objects, and

the evaluation setting unit employs the evaluation function and the determination rule that are suitable for detecting a specific situation predicted for the specified type objects.

9. The image recognition apparatus of claim 8, wherein

the specified type objects are pedestrians, and

the specific situation includes at least one of an abnormal behavior and a sign of the abnormal behavior of the pedestrians.

10. The image recognition apparatus of claim 1 further comprising:

a third evaluation unit in the evaluation value calculation unit for calculating, as one of the multiple evaluation values, motion evaluation values based on a motion vector that is derived from the motion-compensating prediction.

11. The image recognition apparatus of claim 10, wherein

the third evaluation unit calculates, as one of the motion evaluation values, a size of the motion vector.

12. The image recognition apparatus of claim 10, wherein

the third evaluation unit calculates, as one of the motion evaluation values, a similarity between the motion vector and a panning vector that is caused by a camera motion.

13. The image recognition apparatus of claim 10, wherein

the third evaluation unit calculates, as one of the motion evaluation values, a similarity among the motion vectors of adjacent pixel blocks.

14. The image recognition apparatus of claim 13 further comprising:

a fourth evaluation unit the evaluation value calculation unit, wherein,

after setting motion areas that are formed from consecutive pixel blocks respectively having an over-threshold motion vector similarity relative to a predetermined threshold, the fourth evaluation unit calculates, as one of the multiple evaluation values, for each of the motion areas, an area evaluation value indicative of an area size and an area shape of the motion area.

15. The image recognition apparatus of claim 1, wherein

the determination unit determines whether an evaluation represented as a time-series model that is derived from a comparison of a pattern of the evaluation values in a time-series relative to predetermined time-series models is significant or not.

16. The image recognition apparatus of claim 1, wherein

the determination unit determines, by comparing the evaluation value based on the two dimensional DCT coefficient with a predetermined determination threshold, whether the evaluation value is significant or not, and

the determination unit variably sets the determination threshold according to a frame interval between a comparison frame used in the motion-compensating prediction and a current frame.