Method and system for displaying stereo image by cascade structure and analyzing target in image

Patent metadata

Field	Value
Publication number	US-9679205-B2
Application number	US-201514731577-A
Country	US
Kind code	B2
Filing date	Jun 5, 2015
Priority date	Jun 13, 2014
Publication date	Jun 13, 2017
Grant date	Jun 13, 2017

Abstract

A method and a system for analyzing a target in a stereo image by displaying the stereo image using a cascade structure are disclosed. The method includes for the input stereo image, generating, based on a first relevant feature, rule or model of the stereo image, at least a first first-level structure map, each of the first first-level structure maps being generated based on an individual tolerance level of the first relevant feature, rule or model, and each of the first first-level structure maps including the target at an individual first division level; and at least partly integrating the first first-level structure maps and analyzing the target in the stereo image, to obtain a structure map of a first-level target analysis result including the target.

First claim

What is claimed is:

1. A method for analyzing a target in a stereo image by displaying the stereo image using a cascade structure, the method comprising:
generating a plurality of first-level structure maps based on the stereo image, each first-level structure map being based on an individual tolerance level of a first relevant feature, rule or model of the stereo image, each first-level structure map of the plurality of first-level structure maps including a representation of the target divided according to an individual first division level such that the first-level structure maps include separate, respective first representations of the target divided according to separate, respective first division levels; and
at least partly integrating the plurality of first-level structure maps to generate a structure map of a first-level target analysis result including a representation of the target, wherein the at least partly integrating includes performing mutual correction of the plurality of first-level structure maps to determine a division level, of the first division levels, that is associated with a representation of the target, of the representations of the target, having a greatest level of accuracy based on a feature of the target, to generate a structure map of the first-level target analysis result that includes a representation of the target divided according to the determined first division level.

2. The method for analyzing a target according to claim 1 , the method comprising:
generating at least one second first-level structure map based on the stereo image, each second first-level structure map being based on an individual tolerance level of a second relevant feature, rule or model that is different from the first relevant feature, rule or model, each second first-level structure map including a representation of the target divided according to an individual second division level; and
at least partly integrating the plurality of first-level structure maps and the at least one second first-level structure map to generate a structure map of a first-level target analysis result including a representation of the target.

3. The method for analyzing a target according to claim 2 , the method comprising:
generating at least one second-level structure map based on the structure map of the first-level target analysis result, each second-level structure map being based on an individual tolerance level of a third relevant feature, rule or model that is different from the first and second relevant features, rules or models, each second-level structure map including a representation of the target divided according to an individual third fitting level; and
at least partly integrating the at least one second-level structure map to generate a structure map of a second-level target analysis result including a representation of the target.

4. The method for analyzing a target according to claim 3 , wherein,
at least one of the first relevant feature, rule or model and the second relevant feature, rule or model includes a space adjacency feature, a grayscale conformity feature, an edge continuity feature, or a contour feature associated with clustering,
the individual tolerance level of the first relevant feature, rule or model or the individual tolerance level of the second relevant feature, rule or model includes one of,
clustering based on an individual space disparity threshold using a fixed disparity center or a movable disparity center in a case of the space adjacency feature,
clustering based on an individual grayscale difference threshold in a case of a grayscale consistency feature,
clustering based on an individual edge continuity threshold in a case of the edge continuity feature, or
clustering based on an individual contour inclusion relation in a case of the contour feature, and

the third relevant feature, rule or model includes a vertical line fitting model, a plane fitting model or a cube fitting model.

5. The method for analyzing a target according to claim 3 , wherein,
at least partly integrating the at least one second-level structure map to generate the structure map of the second-level target analysis result including the representation of the target includes,
performing mutual correction of the at least one second-level structure map to determine a third fitting level, of at least one third fitting level, that is associated with a representation of the target, of the representations of the target, having a greatest level of accuracy based on a feature of the target, to generate a structure map of the second-level target analysis result that includes a representation of the target divided according to the determined third fitting level.

6. The method for analyzing a target according to claim 5 , wherein,
performing mutual correction of the at least one second-level structure map including the representation of the target divided according to the individual third fitting level includes,
after determining the third fitting level that is associated with a representation of the target having a greatest level of accuracy, performing merging or division for the representations of the targets divided according to the at least one third fitting level to generate the representation of the target divided according to the determined third fitting level.

7. The method for analyzing a target according to claim 2 , wherein,
at least partly integrating the plurality of first-level structure maps and the at least one second first-level structure map to generate the structure map of the first-level target analysis result including the representation of the target includes,
correcting the plurality of first-level structure maps based on the at least one second first-level structure map to determine a division level, of the first division levels and at least one second division level, that is associated with a representation of the target, of the first representations of the target and at least one second representation of the target, having a greatest level of accuracy based on a feature of the target, to generate a structure map of the first-level target analysis result that includes a representation of the target divided according to the determined division level.

8. The method for analyzing a target according to claim 7 , wherein,
correcting the plurality of first-level structure maps based on the at least one second first-level structure map includes,
after determining the division level that is associated with a representation of the target having a greatest level of accuracy, performing merging or division for the representations of the target divided according to the first division levels or the at least one second division level, to generate the representation of the target divided according to the determined division level.

9. The method for analyzing a target according to claim 1 , wherein,
performing mutual correction of the plurality of first-level structure maps including the target at the individual first division level includes,
after determining the division level that is associated with a representation of the target having a greatest level of accuracy, performing merging or division for the representations of the target divided according to the first division levels to generate the representation of the target divided according to the determined division level.

10. The method for analyzing a target according to claim 1 , wherein,
the feature of the target includes one of,
a symmetry of the target,
a horizontal grayscale conformity of the target, or
an inclusion characteristic of an inner contour of the target.

11. The system of claim 10 , the pro

Assignees

• Chen Chao
• Lu Yaojie
• Shi Zhongchao
• Liu Dianchao
• Ricoh Co Ltd

Inventors

• Chen Chao
• Lu Yaojie
• Shi Zhongchao
• Liu Dianchao

Classifications

• G06K9/00805 Primary

  Physics · mapped topic

• G06K2209/21

  Physics · mapped topic

• G06T7/73

  using feature-based methods · CPC title

• G06V20/58 Primary

  Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads · CPC title

• G06V2201/07

  Target detection · CPC title