Image-processing device, surveillance system, method for establishing a scene reference image, and computer program

• US 8,285,044 B2
• Filed: 07/18/2007
• Issued: 10/09/2012
• Est. Priority Date: 07/21/2006
• Status: Active Grant

• Current Assignee: Robert Bosch GmbH
• Original Assignee: Robert Bosch GmbH
• Inventors: Niem, Wolfgang, Merkel, Marcel, Mueller-Scheiders, Stefan
• Primary Examiner(s): Krasnic, Bernard
• Application Number:
  US11/779,433
• Publication Number:
  US 20080019566A1
• Time in Patent Office: 1,910 Days
• Field of Search: 382/173, 382/103
• US Class Current: 382/173
• CPC Class Codes: G06V 20/52 Surveillance or monitoring of activities, e.g. for recognising suspicious objects recognising microscopic objects G06V20/69
• CPC Class Codes: G06V 20/54 of traffic, e.g. cars on the road, trains or boats

Abstract

An image-processing device carries out an object segmentation in which the object segmentation is executed and/or is executable through comparison of a camera image to a scene reference image of a surveillance scene, equipped with a learning device for generating the scene reference image; the learning device generates the scene reference image through evaluation of a medium-term and/or long-term observation of the surveillance scene a) that extends over a time period of longer than one day, preferably longer than several days, in particular longer than 1 week, and/or b) that extends over a time period that includes several states of the surveillance scene.

Claims

1. An image-processing device for carrying out an object segmentation, comprising:

means for executing the object segmentation through comparison of a camera image to a received reference image of a surveillance scene;

a learning device for generating the scene reference image, said learning device being configured to generate the scene reference image through evaluation of a medium-term and/or long-term observation of the surveillance scene that occurs in a situation selected from the group consisting of occurring over a time period of longer than one day, occurring over a time period which includes several states of the reference scene, and both,

wherein said learning device is configured to generate a scene model, wherein the scene model is a model of a scene that comprises artificial model objects in the form of a virtual world, and wherein said scene model is in the form of a multi-mode scene model and individual modes of said multi-mode scene model are distinguished through different states of the surveillance scene, and wherein the states of the surveillance scene differ with regard to one or more of parameters selected from the group consisting of relative position of a surveillance camera, relative orientation of a surveillance camera, absolute position of a surveillance camera, and absolute orientation of the surveillance camera, wherein the scene model forms the basis for the generation of the scene reference image.

2. The image-processing device as defined in claim 1, wherein said learning device is configured so as to generate the scene reference image through the evaluation of a medium-term and/or long-term observation of the surveillance scene that occurs longer than several days.

3. The image-processing device as defined in claim 1, wherein said learning device is configured so as to generate the scene reference image through the evaluation of a medium-term and/or long-term observation of the surveillance scene that occurs longer than one week.

4. The image-processing device as defined in claim 1, wherein said learning device is configured to generate the scene model of the surveillance scene.

5. The image-processing device as defined in claim 4, wherein said learning device is configured so that it generates the scene model including objects selected from the group consisting of trajectory-based objects, image-based objects, and both.

6. The image-processing device as defined in claim 5, wherein said learning device is configured so as to generate the scene model including the trajectory-based objects selected from the group consisting of paths, walkways, streets, entrances, exits, static objects, and combinations thereof.

7. The image-processing device as defined in claim 5, wherein said learning device is configured so as to generate the scene model including the image-based objects selected from the group consisting of interference regions, reflective regions, shaded regions, and combinations thereof.

8. The image-processing device as defined in claim 1, wherein the states of the surveillance scene differ with regard to another one or more of parameters selected further from the group consisting of hour, time of day, date, day of the week, season, and lunar phase.

9. A surveillance system, comprising:

an image processing device for carrying out an object segmentation and including means for executing the object segmentation through comparison of a camera image to a received reference image of a surveillance scene;

a learning device for generating the scene reference image, said learning device being configured to generate the scene reference image through evaluation of a medium-term and/or long-term observation of the surveillance scene that occurs in a situation selected from the group consisting of occurring over a time period of longer than one day, occurring over a time period which includes several states of the reference scene, and both, wherein said learning device is configured to generate a scene model, wherein the scene model is a model of a scene that comprises artificial model objects in the form of a virtual world, and wherein said scene model is in the form of a multi-mode scene model and individual modes of said multi-mode scene model are distinguished through different states of the surveillance scene, and wherein the states of the surveillance scene differ with regard to one or more of parameters selected from the group consisting of relative position of a surveillance camera, relative orientation of a surveillance camera, absolute position of a surveillance camera, and absolute orientation of the surveillance camera, wherein the scene model forms the basis for the generation of the scene reference image;

a plurality of interfaces for connection to surveillance cameras; and

an evaluation device that is configured to establish proximity relationship between said surveillance cameras through use of trajectory-based objects.

10. The surveillance system as defined in claim 9, wherein the trajectory-based objects are implemented in form selected from the group consisting of entrances, exits, and both.

11. A method for generating a scene reference image of a surveillance scene for an object segmentation in image processing, comprising the steps of:

providing a learning device for generating the scene reference image;

generating a scene reference image with said learning device by evaluating a medium-term and/or long-term observation of the surveillance scene that occurs over a time period selected from the group consisting of a time period of longer than one day, over a time period which includes several states of the surveillance scene, and both, wherein said learning device is configured to generate a scene model, wherein the scene model is a model of a scene that comprises artificial model objects in the form of a virtual world, and wherein said scene model is in the form of a multi-mode scene model and individual modes of said multi-mode scene model are distinguished through different states of the surveillance scene, and wherein the states of the surveillance scene differ with regard to one or more of parameters selected from the group consisting of relative position of a surveillance camera, relative orientation of a surveillance camera, absolute position of a surveillance camera, and absolute orientation of the surveillance camera, wherein the scene model forms the basis for the generation of the scene reference image.

12. The method as defined in claim 11, wherein said generating of the scene reference image includes generating the scene reference image over a time period longer than several days.

13. The method as defined in claim 11, wherein said generating of the scene reference image includes generating the scene reference image over a time period longer than one week.

14. A non-transitory computer readable medium containing a computer program with programming code means for executing a method on an image processing device for carrying out an object segmentation, said image processing device including means for executing the object segmentation through comparison of a camera image to a received reference image of a surveillance scene, a learning device for generating the scene reference image, said learning device being configured to generate the scene reference image through evaluation of a medium-term and/or long-term observation of the surveillance scene that occurs in a situation selected from the group consisting of occurring over a time period of longer than one day, occurring over a time period which includes several states of the reference scene, and both when the program is executed on a computer, wherein said learning device is configured to generate a scene model wherein the scene model is a model of a scene that comprises artificial model objects in the form of a virtual world, and wherein said scene model is in the form of a multi-mode scene model and individual modes of said multi-mode scene model are distinguished through different states of the surveillance scene, and wherein the states of the surveillance scene differ with regard to one or more of parameters selected from the group consisting of relative position of a surveillance camera, relative orientation of a surveillance camera, absolute position of a surveillance camera, and absolute orientation of the surveillance camera, wherein the scene model forms the basis for the generation of the scene reference image;

a plurality of interfaces for connection to surveillance cameras; and

an evaluation device that is configured to establish proximity relationship between said surveillance cameras through use of trajectory-based objects.