Method and apparatus for indicating lane

Patent metadata

Field	Value
Publication number	US-2018129887-A1
Application number	US-201715467190-A
Country	US
Kind code	A1
Filing date	Mar 23, 2017
Priority date	Nov 7, 2016
Publication date	May 10, 2018
Grant date	—

Abstract

Disclosed are methods and apparatuses for indicating a lane, the method including detecting a lane marking from an input image, acquiring a bird's eye view image by applying an inverse perspective mapping to the detected lane marking, extracting control points corresponding to the lane marking from the bird's eye view image, and indicating a lane by re-projecting the control points to a space of the input image through an inverse transformation.

First claim

What is claimed is:

1 . A method of indicating a lane, the method comprising:
detecting a lane marking from an input image; acquiring a bird's eye view image by applying an inverse perspective mapping to the detected lane marking; extracting control points corresponding to the lane marking from the bird's eye view image; and indicating a lane by re-projecting the control points to a space of the input image through an inverse transformation.

2 . The method of claim 1 , wherein the detecting of the lane marking comprises detecting the lane marking from the input image using a convolutional neural network (CNN) trained to recognize the lane marking.

3 . The method of claim 1 , wherein the detecting of the lane marking comprises:
dividing a searching area of the input image using a sliding window extending in a horizontal direction; and detecting the lane marking from the divided searching area.

4 . The method of claim 3 , wherein the sliding window comprises a bounding box in a form of a stripe extending in a horizontal direction, and
a size of the stripe is gradually reduced in a direction from bottom to top of the searching area.

5 . The method of claim 4 , wherein the acquiring of the bird's eye view image comprises acquiring the bird's eye view image by applying the inverse perspective mapping based on a center of the bounding box.

6 . The method of claim 2 , wherein the CNN is trained on lane markings of various road surface views.

7 . The method of claim 2 , wherein the CNN is trained to identify a bounding box of a lane marking to be detected from the input image and a type of the lane marking to be detected.

8 . The method of claim 1 , wherein the extracting of the control points comprises performing a curve approximation on the bird's eye view image and extracting the control points corresponding to the lane marking from the bird's eye view image.

9 . The method of claim 8 , wherein the extracting of the control points comprises:
converting the lane marking into a curve by performing the curve approximation on the bird's eye view image; and extracting the control points corresponding to the curve.

10 . The method of claim 1 , wherein the extracting of the control points comprises performing a cubic Bezier curve approximation on the bird's eye view image through a random sample consensus (RANSAC) spline fitting and extracting the control points corresponding to the lane marking.

11 . The method of claim 1 , wherein the extracting of the control points comprises:
determining a straight line tendency of the lane marking by performing a Hough transformation on the bird's eye view image; establishing a bounding box proportional to a lane distance based on the straight line tendency; and extracting the control points corresponding to the lane marking by performing a RANSAC spline fitting on the lane marking in an area of the bounding box.

12 . The method of claim 1 , wherein the input image comprises at least one of a road view or a road surface view.

13 . A non-transitory computer-readable medium storing instructions that, when executed by a processor, cause the processor to perform the method of claim 1 .

14 . An apparatus for indicating a lane, the apparatus comprising:
a communication interface configured to receive an input image; and a processor configured to detect a lane marking from an input image, extract control points corresponding to the lane marking from a bird's eye view image acquired by applying an inverse perspective mapping to the lane marking, re-project the control points to a space of the input image through an inverse transformation, and indicate a lane.

15 . The apparatus of claim 14 , further comprising:
a memory storing parameters of a convolutional neural network (CNN) trained to recognize the lane marking, wherein the processor is further configured to detect the lane marking from the input image using a CNN to which the parameters are applied.

16 . The apparatus of claim 14 , wherein the processor is further configured to divide a searching area of the input image using a sliding window extending in a horizontal direction and to detect the lane marking from the divided searching area, and
the sliding window comprising a bounding box in a form of a stripe extending in a horizontal direction.

17 . The apparatus of claim 15 , wherein the CNN is trained to identify a bounding box of a lane marking to be detected from the input image and a type of the lane marking to be detected.

18 . The apparatus of claim 14 , wherein the processor is further configured to perform a curve approximation on the bird's eye view image, to convert the lane marking into a curve, and to extract the plurality of control points corresponding to the curve.

19 . The apparatus of claim 14 , wherein the processor is further configured to perform a cubic Bezier curve approximation on the bird's eye view image through a random sample consensus (RANSAC) spline fitting and to extract the control points corresponding to the lane marking.

20 . The apparatus of claim 14 , wherein the processor is further configured to perform a Hough transformation on the bird's eye view image, to calculate a straight line tendency of the lane marking, to establish a bounding box proportional to a lane distance based on the straight line tendency, to perform a RANSAC spline fitting on the lane marking in an area of the bounding box, and to extract the plurality of control points corresponding to the lane marking.

Assignees

• Samsung Electronics Co Ltd

Inventors

• Kang Nahyup
• Sung Young Hun
• Lee Keechang

Classifications

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/10032

  Satellite or aerial image; Remote sensing · CPC title

• G06T2207/30256

  Lane; Road marking · CPC title

• G06K9/00798 Primary

  Physics · mapped topic

• G06T2207/20081

  Training; Learning · CPC title