Method for traffic sign recognition

Patent metadata

Field	Value
Publication number	US-9418303-B2
Application number	US-201013388095-A
Country	US
Kind code	B2
Filing date	Sep 24, 2010
Priority date	Oct 1, 2009
Publication date	Aug 16, 2016
Grant date	Aug 16, 2016

Abstract

A traffic sign recognition method analyzes and classifies image data of a sensor in an information processing unit. The image data is analyzed to select an image portion judged to contain a traffic sign of a particular sign class. A class-specific feature is identified in the image portion. A modified image portion is created, in which the class-specific feature has been shifted to a center of the modified image portion. Then the modified image portion is evaluated by a learning-based algorithm to recognize the traffic sign of the particular sign class.

First claim

The invention claimed is:

1. A method of traffic sign recognition that analyzes and classifies image data provided by a sensor in an information processing unit, wherein
an initial image detail that sufficiently probably contains an image of an entire actual traffic sign belonging to a particular class of traffic signs is determined from results of an analysis of the image data,
a class-specific feature of the entire actual traffic sign is identified in the initial image detail at a location away from a geometric center of the entire actual traffic sign relative to an outer border of the entire actual traffic sign that extends around an outer perimeter of the entire actual traffic sign,
the initial image detail is modified by shifting the class-specific feature so as to thereby produce a modified image detail that contains an image of an entire modified traffic sign, which corresponds to the image of the entire actual traffic sign except that the class-specific feature has been shifted to a geometric center of the entire modified traffic sign relative to the outer border thereof in the modified image detail and thereby replaces what is at the geometric center of the entire actual traffic sign in the initial image detail, and except that at least one adjacent image region is padded with filler pixels, wherein the at least one adjacent image region is adjacent to the class-specific feature in the entire modified traffic sign in the modified image detail, and
the modified image detail is submitted to a classificator which evaluates the image of the entire modified traffic sign with the class-specific feature at the geometric center of the entire modified traffic sign in the modified image detail by a learning-based algorithm to recognize the entire modified traffic sign.

2. The method according to claim 1 , characterized in that the class-specific feature is further adjusted to a predetermined image size in the modified image detail.

3. The method according to claim 1 , characterized in that the initial image detail is additionally evaluated as to a size thereof, and the modified image detail is produced only if the initial image detail has a predetermined size.

4. The method according to claim 1 , characterized in that the initial image detail is submitted to the classificator which attempts to recognize the entire actual traffic sign from the initial image detail, and the modified image detail is produced only if the attempt to recognize the entire actual traffic sign from the initial image detail is unsuccessful.

5. The method according to claim 1 , characterized in that the filler pixels correspond to other pixels of surroundings of the class-specific feature.

6. The method according to claim 1 , characterized in that the filler pixels are calculated from other pixels of surroundings of the class-specific feature.

7. The method according to claim 1 , characterized in that the particular class of traffic signs comprises circular traffic signs, and the entire actual traffic sign is a circular traffic sign of which the outer border comprises a contrasting circular border line around the outer perimeter thereof.

8. The method according to claim 1 , characterized in that the particular class of traffic signs comprises traffic signs with a numerical block as the class-specific feature.

9. A method of traffic sign recognition, comprising steps:
a) providing image data from an image sensor;
b) in an information processing arrangement, analyzing the image data and selecting an initial image portion thereof that includes an image of an entire actual traffic sign that is bounded by an outer border that extends around an outer perimeter of the entire actual traffic sign;
c) in the information processing arrangement, evaluating the image of the entire actual traffic sign and therein identifying an informational block of the entire actual traffic sign at a geometrically uncentered position relative to the outer border;
d) in the information processing arrangement, shifting a position of the informational block to a geometric center relative to the outer border within the image of the entire actual traffic sign to produce therefrom a modified image portion containing an image of an entire modified traffic sign in which the informational block is positioned at the geometric center of the entire modified traffic sign in the modified image portion; and
e) in the information processing arrangement, evaluating the image of the entire modified traffic sign in the modified image portion to recognize the entire modified traffic sign including the informational block at the geometric center thereof as only a portion thereof.

10. The method according to claim 9 , wherein the outer border comprises an outer peripheral portion of the entire actual traffic sign in the image of the entire actual traffic sign.

11. The method according to claim 9 , wherein the entire actual traffic sign includes a background field on which the informational block is disposed, and wherein the outer border comprises a contrasting border line extending around the background field.

12. The method according to claim 9 , wherein the informational block is a numerical block that contains only numbers.

13. The method according to claim 9 , wherein the shifting of the position of the informational block in the step d) creates an area that had been occupied by the informational block in the initial image portion and that is adjacent to the informational block in the modified image portion, and further comprising filling the area with filler pixels determined from a background field of the entire actual traffic sign on which the informational block is disposed.

14. The method according to claim 13 , further comprising overwriting with additional filler pixels an additional area that is located adjacent to the informational block in the modified image portion and that contains informational symbols other than and smaller than the informational block.

15. The method according to claim 9 , wherein the shifting of the position of the informational block comprises copying or shifting pixels of the informational block.

16. The method according to claim 9 , wherein the evaluating of the image of the entire modified traffic sign in the modified image portion to recognize the entire modified traffic sign is performed according to a learning-based algorithm.

17. A method of traffic sign recognition, comprising steps:
a) with an image sensor, capturing an initial image of an entire actual traffic sign that is bounded around an outer perimeter thereof by an outer border of the entire actual traffic sign, and that includes a sign background, and an information block at a geometrically uncentered position relative to the outer border;
b) in an information processing arrangement, analyzing the initial image to detect the entire actual traffic sign and the information block;
c) in the information processing arrangement, producing a copy of the information block and overwriting the copy of the information block onto the initial image of the entire actual traffic sign at a geometrically centered position relative to the outer border so as to thereby produce a modified image of an entire modified traffic sign that includes the outer border, the sign background, and the copy of the information block except at least that the copy of the information block in the modified image is at the geometrically centered position relative to the outer border of the entire modified traffic sign;
d) in the information processing arrangement, evaluating the entire modified

Assignees

• Zobel Matthias
• Conti Temic Microelectronic Gmbh

Inventors

• Zobel Matthias

Classifications

• G06V10/32

  Normalisation of the pattern dimensions · CPC title

• G06K9/42

  Physics · mapped topic

• G06K9/00818 Primary

  Physics · mapped topic

• G06V20/582 Primary

  of traffic signs · CPC title