Detecting geographic features in images based on invariant components

• US 8,761,435 B2
• Filed: 06/24/2009
• Issued: 06/24/2014
• Est. Priority Date: 06/24/2009
• Status: Active Grant

• Current Assignee: HERE Global B.V. (Nokia Corporation)
• Original Assignee: Navteq B.V.
• Inventors: Chen, Xin
• Primary Examiner(s): Repko, Jason M
• Assistant Examiner(s): FUJITA, KATRINA R
• Application Number:
  US12/490,555
• Publication Number:
  US 20100329504A1
• Time in Patent Office: 1,826 Days
• Field of Search: 382/103, 382/104, 382/201, 382/229, 348/116, 348/118
• US Class Current: 382/103
• CPC Class Codes: G06V 20/582 of traffic signs
• CPC Class Codes: G06V 20/588 Recognition of the road, e.g. of lane markings; Recognition of the vehicle driving pattern in relation to the road
• CPC Class Codes: G06V 20/63 Scene text, e.g. street names
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/18143 Extracting features based on salient regional features, e.g. scale invariant feature transform [SIFT] keypoints

Abstract

Systems, devices, features, and methods for detecting geographic features in images, such as, for example, to develop a navigation database are disclosed. For example, a method of detecting text from collected images includes collecting a plurality of images of geographic areas along a road or path. An image of the plurality of images is selected. Components that represent a feature about the road or path in the selected image are determined. In one embodiment, the components are independent or invariant to scale of the feature. The determined components are compared to reference components in a data library. If the determined components substantially match with the reference components, the feature in the selected image is identified to be a text character (e.g., of a road sign) corresponding to at least some of the reference components in the data library.

Claims

1. A non-transitory computer-readable medium configured to store a computer program that performs a method of detecting a road sign in an image for development of a navigation database, the method comprising:

receiving a plurality of images of roads, the plurality of images captured via a camera mounted on a vehicle;

determining gradient vectors of one or more features in each of the plurality of images, the gradient vectors being substantially invariant to scale, rotation, and illumination of the one or more features, the gradient vectors determined as a function of a scale-invariant feature transform;

comparing the determined gradient vectors with reference gradient vectors to determine whether a road sign is included in at least one of the plurality of images, the reference gradient vectors corresponding to a text character;

comparing distances between the determined gradient vectors to identify a cluster of features corresponding to a series of text including the text character; and

identifying at least one of the plurality of images determined to include the road sign as a function of the series of text determined from the gradient vectors of the cluster of features, the reference gradient vectors, and a dictionary of possible road sign text.

2. The method of claim 1, wherein the road sign comprises a road sign with an uncommon text sequence.

3. The method of claim 1, wherein the determined gradient vectors comprise amplitude extrema.

4. The method of claim 1, wherein identifying the at least one of the plurality of images includes generating a text file, the text file including location information of where the identified image was taken.

5. The method of claim 1, wherein determining comprises determining gradient vectors for individual text characters in each of the one or more features.

6. The method of claim 1, wherein identifying comprises identifying when, based on the comparison, the determined gradient vectors substantially match with the reference gradient vectors.

7. A method of detecting text from collected road images to develop a map navigation database, the method comprising:

collecting a plurality of images of geographic areas along a road;

selecting an image of the plurality of images;

determining components comprising gradient vectors that represent one or more features about the road in the selected image, the components being independent to scale of the feature;

performing a first comparison between distances between the components in the image to identify a cluster of components;

performing a second comparison of the cluster of components to reference components in a data library;

when, based on the second comparison, the duster of components substantially match with the reference components, identifying the feature in the selected image to be at least one text character of a series of text corresponding to the reference components in the data library; and

developing a map navigation database based on the series of text in the selected image including the at least one text character.

8. The method of claim 7, wherein developing the map navigation database includes storing data representing a geographic object associated with the identified at least one text character in the map navigation database.

9. The method of claim 7, wherein the determined components are substantially independent to brightness of the feature or substantially independent to rotation of the feature.

10. The method of claim 7, further comprising:

identifying a letter or number in a reference image of the plurality of images;

determining the reference components that represent the letter or number in the reference image, the reference components being independent to scale and to rotation of the letter or number; and

storing the reference components in the data library.

11. The method of claim 7, wherein the distances between the components correspond to a boundary corresponding to individual text characters.

12. The method of claim 7, further comprising:

filtering out components that are more than a predetermined distance from other components.

13. The method of claim 7, further comprising:

filtering out a plurality of components that are within a predetermined distance to each other but are more than the predetermined distance to other components.

14. The method of claim 7, wherein the first comparison is configured to identify successive individual text characters.

15. A method of identifying text in an image to determine a path feature, the method comprising:

identifying a letter or number in an image of a first path using gradient vectors of the image;

determining data components that correspond to the identified letter or number, the data components being scale-invariant components;

storing the data components that correspond to the letter or number in association with a series of text in a navigation library of a map database developer according to a geographic position of the first path;

selecting another image representing a second path;

determining scale-invariant components of the other image; and

comparing the determined scale-invariant components of the other image with the data components stored in the navigation library as a function of a difference between a geographic position of the second path and the geographic position of the first path to determine when the other image includes a character corresponding to the series of text.

16. The method of claim 15, further comprising:

when it is determined that the other image includes the character, identifying the other image as including a path sign.

17. The method of claim 15, further comprising:

clustering some of the determined scale-invariant components of the other image; and

before comparing, filtering determined scale-invariant components that are not part of the clustered components.

18. The method of claim 15, wherein determining the scale-invariant components of the letter or the number comprises determining amplitude extrema in regions of a difference image, the difference image comprising an image generated from a difference between the image of the first path and a blurred image of the first path.

19. The method of claim 15, wherein the data components are invariant to rotation and substantially invariant to illumination.

20. The method of claim 15, further comprising:

when it is determined that the other image includes the character, providing identification of the selected other image to the map database developer.

21. The method of claim 20, wherein providing the identification comprises providing a text file including location information associated with the other image.

22. The method of claim 15, wherein the data components are substantially invariant to rotation, scale, and illumination.