Method Of Generating A Training Image For An Automated Vehicle Object Recognition System

Patent metadata

Field	Value
Publication number	US-2016162750-A1
Application number	US-201514952025-A
Country	US
Kind code	A1
Filing date	Nov 25, 2015
Priority date	Dec 5, 2014
Publication date	Jun 9, 2016
Grant date	—

Abstract

In a method of generating a training image for teaching of a camera-based object recognition system suitable for use on an automated vehicle which shows an object to be recognized in a natural object environment, the training image is generated as a synthetic image by a combination of a base image taken by a camera and of a template image in that a structural feature is obtained from the base image and is replaced with a structural feature obtained from the template image by means of a shift-map algorithm.

First claim

1 . A method of generating a training image ( 10 ) which is in particular provided for the teaching of a camera-based object recognition system and which shows an object ( 20 ′) to be recognized in a natural object environment ( 25 ), wherein the training image ( 10 ) is generated as a synthetic image by combination of a base image ( 11 ) taken by a camera and of a template image ( 13 ) in that a structural feature is removed from the base image ( 11 ) and is replaced with a structural feature obtained from the template image ( 13 ) by means of a shift-map algorithm, wherein the replacement of the structural feature is only carried out by rearrangement of pixels of the base image ( 11 ).

2 . A method in accordance with claim 1 ,
characterized in that the rearrangement of pixels of the base image ( 11 ) is restricted to a part region of the base image ( 11 ).

3 . A method in accordance with claim 1 ,
characterized in that the rearrangement of pixels of the base image ( 11 ) is restricted to a central part region of the base image ( 11 ).

4 . A method in accordance with any one of the preceding claims,
characterized in that optimization criteria are made use of for the shift-map algorithm which comprise a maintenance of proximity relationships of pixels of the base image ( 11 ), an avoidance of tonal value discontinuities in the generated training image ( 10 ) and/or the maintenance of a similarity between the base image ( 11 ) and the template image ( 13 ).

5 . A method in accordance with any one of the preceding claims,
characterized in that the base image ( 11 ) and the template image ( 13 ) are compared with one another to determine a distance dimension indicating the similarity of the images; and in that the distance dimension is used as an optimization criterion for the shift-map algorithm.

6 . A method in accordance with claim 5 ,
characterized in that both the base image ( 11 ) and the template image ( 13 ) are transformed into a canonical reference framework ( 21 ) before they are compared with one another using the distance dimension.

7 . A method in accordance with any one of the preceding claims,
characterized in that the structural feature comprises a texture and/or a pattern.

8 . A method in accordance with any one of the preceding claims,
characterized in that the template image ( 13 ) is an image taken by a camera.

9 . A method in accordance with any one of the claims 1 to 7 ,
characterized in that
the template image ( 13 ) is a graphically generated image.

10 . A method in accordance with any one of the preceding claims,
characterized in that the base image ( 11 ) and the training image ( 10 ) each show a road sign ( 20 , 20 ′) in a road environment ( 25 ).

11 . A method of teaching an object recognition system, in particular a road sign recognition system,
in which training images ( 10 ) are provided which show objects ( 20 , 20 ′) to be recognized in a natural object environment ( 25 ) and in which a recognition algorithm for recognizing the objects ( 20 , 20 ′) is developed or adapted by means of an image processing system using the training images ( 10 ), characterized in that at least one training image ( 10 ) is generated by a method in accordance with any one of the preceding claims

12 . A method in accordance with claim 11 , wherein the recognition algorithm comprises a classification algorithm for associating a recognized object ( 20 , 20 ′) with one of a plurality of predefined object classes.

13 . A computer program product which contains program instructions which execute a method in accordance with any one of the claims 1 to 10 when the computer program is run on a computer.

14 . A computer program product which contains program instructions which execute a method in accordance with one of the claim 11 or 12 when the computer program is run on a computer.

Assignees

• Delphi Tech Inc

Inventors

• Haselhoff Anslem
• Mueller Dennis
• Meuter Mirko
• Nunn Christian

Classifications

• G06V20/582 Primary

  of traffic signs · CPC title

• G06F18/24 Primary

  Classification techniques · CPC title

• G06F18/22

  Matching criteria, e.g. proximity measures · CPC title

• H04N7/18

  Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast · CPC title

• G06K9/46 Primary

  Physics · mapped topic