Systems and methods for lane detection

Patent metadata

Field	Value
Publication number	US-12087010-B2
Application number	US-202117447421-A
Country	US
Kind code	B2
Filing date	Sep 11, 2021
Priority date	Mar 12, 2019
Publication date	Sep 10, 2024
Grant date	Sep 10, 2024

Abstract

Systems and methods for lane detection are provided. The methods may include obtaining an image and determining, for each of a plurality of pixels in an image, a semantic category using a trained semantic segmentation network. The methods may further include determining one or more pixel sets based on the plurality of pixels according to a predetermined rule. Each of the one or more pixel sets includes one or more pixels of a same semantic category. The methods may further include, in response to a determination that there are one or more erroneous pixels in a pixel set, removing the one or more erroneous pixels from the pixel set to obtain a fitting line corresponding to a lane line. The methods may further include determining a position of the lane line based on the fitting line.

First claim

What is claimed is:

1. A method for lane detection, implemented on a computing device including at least one processor and at least one storage medium, the method comprising:
obtaining an image;
determining, for each of a plurality of pixels in the image, a semantic category using a trained semantic segmentation network;
determining one or more pixel sets based on the plurality of pixels according to a predetermined rule, wherein each of the one or more pixel sets includes one or more pixels of a same semantic category; and
for each of the one or more pixel sets,
determining whether there are one or more erroneous pixels that satisfy a predetermined condition by performing an iterative fitting operation on the one or more pixels in the pixel set, including:
in each iteration of the iterative fitting operation, for each of the one or more pixels in the pixel set,
determining, based on an estimated fitting line associated with the one or more pixels in the pixel set, a fitted coordinate associated with the pixel;
determining a first difference between the fitted coordinate and an actual coordinate associated with the pixel; and
in response to a determination that the first difference is greater than a first threshold, determining that the pixel is an erroneous pixel;

in response to a determination that there are one or more erroneous pixels, removing the one or more erroneous pixels from the pixel set to obtain a fitting line corresponding to a lane line, and
determining a position of the lane line based on the fitting line.

2. The method of claim 1 , wherein the determining whether there are one or more erroneous pixels that satisfy a predetermined condition further includes:
in the each iteration,
performing a fitting operation on coordinates of the one or more pixels in the pixel set to obtain the estimated fitting line; and
determining whether there are one or more erroneous pixels that satisfy the predetermined condition.

3. The method of claim 1 , wherein a termination criterion for the iterative fitting operation is that for each of one or more fitting parameters of the fitting line, a second difference between the fitting parameters estimated in two consecutive iterations is less than or equal to a second threshold.

4. The method of claim 1 , further comprising:
performing a binarization operation to obtain binary information of each of the plurality of pixels; and
for each of the one or more pixel sets,
determining, based at least on the fitting line and the binary information of the one or more pixels in the pixel set, a lane line category of the lane line.

5. The method of claim 1 , wherein the trained semantic segmentation network includes one or more convolutional layers and one or more deconvolutional layers, the one or more convolutional layers being configured for a depth-wise separable convolution.

6. A method for lane detection, implemented on a computing device including at least one processor and at least one storage medium, the method comprising:
obtaining an image;
determining, for each of a plurality of pixels in the image, a semantic category using a trained semantic segmentation network;
determining one or more pixel sets based on the plurality of pixels according to a predetermined rule, wherein each of the one or more pixel sets includes one or more pixels of a same semantic category;
performing a binarization operation to obtain binary information of each of the plurality of pixels; and
for each of the one or more pixel sets,
performing a fitting operation on the one or more pixels in the pixel set to obtain a fitting line corresponding to a lane line;
determining a position of the lane line based on the fitting line; and
determining, based at least on the fitting line and the binary information of the one or more pixels in the pixel set, a lane line category of the lane line, including:
for each of the one or more pixels in the pixel set, determining a fitted coordinate of the pixel;
determining a searching range for the pixel based on the fitted coordinate and a lane line width; and
determining the lane line category of the lane line based on one or more searching ranges corresponding to the one or more pixels in the pixel set and the binary information of the one or more pixels in the pixel set.

7. The method of claim 6 , wherein the performing a binarization operation to obtain binary information of each of the plurality of pixels includes:
performing an edge extraction operation on the plurality of pixels in the image to obtain edge information associated with each of the plurality of pixels; and
determining, based on the edge information, the binary information of each of the plurality of pixels.

8. The method of claim 6 , wherein the determining the lane line category of the lane line based on one or more searching ranges corresponding to the one or more pixels in the pixel set and the binary information of the one or more pixels in the pixel set includes:
determining a reference number count of reference pixels, wherein there are no edge pixels near each of the reference pixels within the searching range; and
determining the lane line category of the lane line by comparing the reference number count of pixels and a count threshold.

9. The method of claim 8 , wherein the determining the lane line category of the lane line by comparing the reference number count of pixels and a count threshold includes:
in response to a determination that the reference number count of reference pixels is greater than the count threshold, determining that the lane line category of the lane line is a dashed line; and
in response to a determination that the reference number count of pixels is less than or equal to the count threshold, determining that the lane line category of the lane line is a solid line.

10. The method of claim 6 , further comprising:
determining whether there are at least two pixel sets of the same semantic category in the one or more pixel sets; and
in response to a determination that there are at least two pixel sets of the same semantic category in the one or more pixel sets,
determining a difference between fitting parameters of the at least two pixel sets;
comparing the difference with a parameter difference threshold; and
in response to a determination that the difference is less than or equal to the parameter difference threshold, generating a combined pixel set based on the at least two pixel sets.

11. The method of claim 6 , wherein the fitting operation is an iterative fitting operation, and the method further comprises:
for each of the one or more pixel sets,
determining whether there are one or more erroneous pixels that satisfy a predetermined condition; and
in response to a determination that there are one or more erroneous pixels, removing the one or more erroneous pixels from the pixel set to obtain the fitting line.

12. The method of claim 6 , wherein the trained semantic segmentation network includes one or more convolutional layers and one or more deconvolutional layers, the one or more convolutional layers being configured for a depth-wise separable convolution.

13. A method for lane detection, implemented on a computing device including at least one processor and at least one storage medium, the method comprising:
obtaining an image;
determining, for each of a plurality of pixels in the image, a semantic category using a trained semantic segmentation network, wherein the trained semantic segmentation network include one or more convolutional layers and one or more deconvolutional layers, the one or more convolutional layers being conf

Assignees

• Zhejiang Dahua Technology Co

Inventors

• Zhou Zhangwei
• Li Qiankun
• Lu Wei
• Yin Jun
• Zhang Xingming

Classifications

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06V20/588

  Recognition of the road, e.g. of lane markings; Recognition of the vehicle driving pattern in relation to the road · CPC title

• G06T7/13

  Edge detection · CPC title

• G06T7/12

  Edge-based segmentation · CPC title