Method and apparatus for locating an object of interest within an image

• US 20080004776A1
• Filed: 09/04/2007
• Published: 01/03/2008
• Est. Priority Date: 01/13/2005
• Status: Active Grant

• Current Assignee: TRW Automotive USLLC (Zeppelin - Stiftung der Stadt Friedrichshafen)
• Original Assignee: TRW Automotive USLLC (Zeppelin - Stiftung der Stadt Friedrichshafen)
• Inventors: Wallace, Jon, Luo, Yun
• Granted Patent: US 7,516,005 B2
• Time in Patent Office: Days
• US Class Current: 701/45
• CPC Class Codes: B60R 21/01538 for image processing, e.g. cameras or sensor arrays
• CPC Class Codes: G06T 2207/10012 Stereo images
• CPC Class Codes: G06T 2207/30268 Vehicle interior
• CPC Class Codes: G06T 7/73 using feature-based methods
• CPC Class Codes: G06V 40/103 Static body considered as a whole, e.g. static pedestrian or occupant recognition

Abstract

Systems and methods are provided for identifying a position of a head of a vehicle occupant from a plurality of images of a vehicle interior. Image data is translated from the plurality of images to produce a virtual side view image of the vehicle interior. A largest grouping of pixels in the side view image having values within a desired range is identified. The grouping includes a plurality of pixel columns. A contour comprising uppermost pixels of each pixel column is produced and peaks are located within the contour. The contour peaks are analyzed according to a heuristic voting algorithm to identify a peak location associated with the head of the vehicle occupant.

Claims

1. A method for identifying a position of an object of interest from a plurality of images, comprising the steps of:

obtaining at least one image from each of a plurality of cameras, each camera having an associated perspective of a subject of interest;

translating image data from the plurality of images to produce at least one virtual image of the vehicle interior, each virtual image having a desired perspective different from that of the associated perspectives of the plurality of cameras;

locating candidate objects within the at least one virtual image; and

analyzing the candidate objects according to a heuristic voting algorithm to identify the position of the object of interest.

2. The method of claim 1, wherein one of the at least one virtual images is a side view virtual image having a side view perspective of the subject of interest.

3. The method of claim 2, the step of locating candidate objects within the at least one virtual image comprising identifying a largest grouping of pixels in the side view virtual image having values within a desired range, the grouping comprising a plurality of pixel columns.

4. The method of claim 3, the step of locating candidate objects within the at least one virtual image further comprising:

generating a contour comprising uppermost pixels of each pixel column; and

locating peaks within the contour.

5. The method of claim 4, wherein the step of locating peaks within the contour comprises low pass filtering the generated contour with a low pass filter having a width approximating the expected width of an adult human head within the image.

6. The method of claim 1, further comprising locating an approximate location for a front edge associated with the identified position of the candidate object.

7. The method of claim 6, further comprising iteratively refining the location of the front edge until a desired degree of accuracy is achieved.

8. The method of claim 1, further comprising controlling an occupant protection system according to the identified position of object of interest.

9. The method of claim 1, wherein the step of translating image data to produce a virtual side view image comprises:

constructing a disparity image from the plurality of images;

constructing a surface model of the subject of interest in a global coordinate system;

transforming the surface model into a camera coordinate system associated with the desired perspective; and

projecting the surface model onto an image plane associated with the desired perspective.

10. The method of claim 9, transforming the surface model into a camera coordinate system associated with the desired perspective comprises mapping individual pixels in the surface model to a multiple pixel pattern within the virtual side view image.

11. The method of claim 1, wherein one of the at least one virtual images is a front view virtual image having a front view perspective of the subject of interest.

12. The method of claim 1, further comprising the steps of:

segmenting a portion of the virtual image representing the object of interest from the virtual image; and

providing the segmented portion of the virtual image to a pattern recognition classifier.

13-24. -24. (canceled)