Corresponding point searching method and distance detection device

Patent metadata

Field	Value
Publication number	US-2016358325-A1
Application number	US-201615171606-A
Country	US
Kind code	A1
Filing date	Jun 2, 2016
Priority date	Jun 4, 2015
Publication date	Dec 8, 2016
Grant date	—

Abstract

A corresponding point searching method searches corresponding points in plural images, acquired by in-vehicle cameras, for each pixel in a reference image by using a predetermined first method, for example, the Viterbi algorithm. The method searches corresponding points in the plural images for each pixel in the reference image by using a predetermined second method, for example, an optical flow method. The method detects whether or not a search accuracy of the corresponding points in each region divided in the reference image obtained by the predetermined first method is not less than a reference value. When not less than the reference value, the method selects the corresponding points obtained by the predetermined first method. When less than the reference value, the searching method selects the corresponding points obtained by the predetermined second method. The searching method provides the corresponding points between the plural images with a high accuracy.

First claim

What is claimed is:

1 . A corresponding point searching method for searching for corresponding points between plural images, performed by an image processing device capable of performing an image processing on the plural images having a parallax therebetween, the method comprising:
a first searching step of performing a predetermined first method for searching for corresponding points between the plural images for each pixel in a reference image in the plural images; a second searching step of performing a predetermined second method for searching the corresponding points between the plural images for each pixel in the reference image in the plural images, the predetermined second method being different from the predetermine first method; a search accuracy judgment step of detecting whether or not search accuracy of the corresponding points obtained by the predetermined first method is not less than a reference value; and a corresponding point setting step of selecting the corresponding points obtained by the predetermined first method when the search accuracy of the corresponding points obtained by the predetermined first method is not less than the reference value, and selecting the corresponding points obtained by the predetermined second method when the search accuracy of corresponding points obtained by the predetermined first method is less than the reference value.

2 . The corresponding point searching method according to claim 1 , wherein the corresponding point setting step firstly sets the corresponding points between the images obtained by the predetermined first method, and replaces the corresponding points in a region having search accuracy of less than the reference value with the corresponding points in the region obtained by the predetermined second method.

3 . The corresponding point searching method according to claim 1 , wherein the first searching step comprises:
a pixel cost calculation step calculating a pixel cost for each reference pixel in the reference image on a basis of a difference between reference pixel information representing reference pixels in the reference image and comparison pixel information representing comparison pixels in the reference image while exchanging the reference pixel and the comparison pixels, where the comparison image is selected from the plural images other than the reference image; a minimum cost pixel calculation step calculating a parallax cost for each reference pixels, where the parallax cost represents a parallax change amount as a coordinate difference between the reference pixel and the comparison pixel when the reference pixel is changed, and calculates a combination of the reference pixel and the comparison pixel with a total minimum cost of the pixel cost and the parallax cost; and a corresponding point extraction step extracting the comparison pixel as a corresponding pixel candidate which corresponds to the reference pixel for each reference pixel, wherein the minimum cost pixel calculation step calculates a minimum cost when the reference pixels and the comparison pixels are changed in each of plural directions, and calculates the total minimum cost by adding the minimum costs obtained in the plural directions together.

4 . The corresponding point searching method according to claim 2 , wherein the first searching step comprises:
a pixel cost calculation step calculating a pixel cost for each reference pixel in the reference image on a basis of a difference between reference pixel information representing reference pixels in the reference image and comparison pixel information representing comparison pixels in the reference image while exchanging the reference pixel and the comparison pixels, where the comparison image is selected from the plural images other than the reference image; a minimum cost pixel calculation step calculating a parallax cost for each reference pixels, where the parallax cost represents a parallax change amount as a coordinate difference between the reference pixel and the comparison pixel when the reference pixel is changed, and calculates a combination of the reference pixel and the comparison pixel with a total minimum cost of the pixel cost and the parallax cost; and a corresponding point extraction step extracting the comparison pixel as a corresponding pixel candidate which corresponds to the reference pixel for each reference pixel, wherein the minimum cost pixel calculation step calculates a minimum cost when the reference pixels and the comparison pixels are changed in each of plural directions, and calculates the total minimum cost by adding the minimum costs obtained in the plural directions together.

5 . The corresponding point searching method according to claim 1 , wherein the second searching step extracts the corresponding pixel candidate in the comparison images which corresponds to each pixel in the reference image by using an optical flow.

6 . The corresponding point searching method according to claim 2 , wherein the second searching step extracts the corresponding pixel candidate in the comparison images which corresponds to each pixel in the reference image by using an optical flow.

7 . A distance detection device capable of detecting a distance to an object by using plural acquired images having a parallax, comprising:
a pixel corresponding point section obtaining a correspondence in location between pixels in the plural acquired images on the basis of the corresponding point searching method according to claim 1 ; and a detection section detecting a distance to an object on the basis of the correspondence in location between pixels in the plural acquired images.

8 . A distance detection device capable of detecting a distance to an object by using plural acquired images having a parallax, comprising:
a pixel corresponding point section obtaining a correspondence in location between pixels in the plural acquired images on the basis of the corresponding point searching method according to claim 2 ; and a detection section detecting a distance to an object on the basis of the correspondence in location between pixels in the plural acquired images.

Assignees

• Denso Corp
• Toyota School Found

Inventors

• Ishimaru Kazuhisa
• Shirai Noriaki
• Tehrani Niknejad Hossein
• Mita Seiichi
• Xie Qiwei
• Long Qian

Classifications

• G06T7/593 Primary

  from stereo images · CPC title

• G06F18/22

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

• G06K9/6215

  Physics · mapped topic

• G06K9/4604

  Physics · mapped topic

• G06K9/6202

  Physics · mapped topic