Method for lane and roadway detection

Patent metadata

Field	Value
Publication number	US-11443530-B2
Application number	US-201917053882-A
Country	US
Kind code	B2
Filing date	May 6, 2019
Priority date	May 9, 2018
Publication date	Sep 13, 2022
Grant date	Sep 13, 2022

Abstract

A method for lane and roadway detection uses a multitask DNN architecture including an encoder, a first decoder and a second decoder. The method includes the following steps:providing an input image by an optical detection device,filtering the input image by the encoder,generating a first representation of the lane and/or roadway by the encoder,processing the first representation in the first and second decoders,outputting two different representations from the first and second decoders,combining the two different representations, andoutputting identified lanes, lane markings and the roadway.

First claim

The invention claimed is:

1. A method using a multitask DNN architecture that includes an encoder, a first decoder and a second decoder, wherein the method comprises:
providing an image of a roadway by an optical detection device;
filtering the image by the encoder to produce an encoder output;
providing the encoder output to the first decoder and the second decoder;
in the first decoder, performing a semantic segmentation on the encoder output to produce a pixel classification of the image;
in the second decoder, performing a keypoint regression on the encoder output to produce keypoints of the image;
combining the pixel classification and the keypoints by fusion thereof to produce fused data;
processing the fused data to detect the roadway, lane markings on the roadway, and a lane of the roadway bounded by the lane markings; and
outputting roadway data identifying the roadway, the lane markings and the lanes.

2. The method according to claim 1 , further comprising operating a driver assistance system of a vehicle in response to and dependent on the roadway data.

3. The method according to claim 1 , wherein the encoder output supplies learned features as convolution filters.

4. The method according to claim 1 , wherein the pixel classification produced by the first decoder supplies a pixel-precise classification of a drivable region within the image.

5. The method according to claim 4 , wherein the drivable region is the roadway and/or the lane of the roadway.

6. The method according to claim 1 , wherein the keypoints produced by the second decoder supply visible boundaries of the lane and/or of the roadway as continual pairs of values in image coordinates.

7. The method according to claim 1 , further comprising assessing an identification confidence for the lane and/or for the roadway based on the pixel classification produced by the first decoder and the keypoints produced by the second decoder.

8. The method according to claim 1 , wherein the pixel classification produced by the first decoder defines a free area of the lane or of the roadway, and the keypoints produced by the second decoder define visible portions of boundaries of the lane or of the roadway.

9. The method according to claim 8 , further comprising interpolating concealed portions of the boundaries based on the keypoints.

Assignees

• Conti Temic Microelectronic Gmbh

Inventors

• Damerow Florian
• Fischer Andreea Mirela
• Linsenmann Christopher
• Ide Kai
• Limmer Matthias
• Hegemann Stefan

Classifications

• G06N3/0455

  Auto-encoder networks; Encoder-decoder networks · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/08

  Learning methods · CPC title

• G06T7/12

  Edge-based segmentation · CPC title

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title