Lane Line Processing via Connected Domain Analysis

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional lane line detection methods in automatic driving are susceptible to image quality and noise, requiring cumbersome post-processing schemes that are not robust.

Innovation Solution

A lane line processing method involving deep learning, binarization processing, connected domain analysis, and polynomial fitting to accurately identify and represent lane lines, which are not affected by image quality and provide robust results.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional image processing methods are used for lane line detection, then the detection process can be implemented, but the results are susceptible to noise points and image quality, reducing robustness

Engineering Contradiction:
Improverobustness of lane line detectionVSAvoidsusceptibility to noise points and image quality
Core Design Contradiction:
ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The patent segments the lane line detection process into distinct modules: a deep learning-based pixel classification module that identifies lane line pixels, and a post-processing module that performs connected component analysis and curve fitting. This segmentation allows each module to specialize in specific tasks, improving overall robustness by isolating noise-sensitive operations from noise-critical operations.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary processing step between pixel classification and final lane line extraction. The connected component analysis serves as an intermediary that filters and organizes classified pixels into coherent lane line structures before curve fitting, reducing the impact of noise points on the final result.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If deep learning methods are used for lane line detection, then detection accuracy can be improved, but cumbersome post-processing schemes are still required

Engineering Contradiction:
Improvelane line detection accuracyVSAvoidcomplexity of post-processing schemes
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts and eliminates unnecessary post-processing steps from the deep learning pipeline. By directly using the pixel classification output from the deep learning model and applying only essential processing (connected component analysis and polynomial fitting), the system removes cumbersome intermediate steps while preserving detection accuracy.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The deep learning model is designed to output pixel-level classification results that directly enable lane line extraction without requiring complex external post-processing. The model serves its own output requirements by providing sufficiently accurate pixel classifications that can be processed through simple, efficient algorithms.

Inventive Principle:
Principle #25Self-service

3Ease of operation

If Hough transform is used to derive lane lines, then lane line extraction can be achieved, but the method is not robust due to susceptibility to noise points

Engineering Contradiction:
Improveease of lane line extractionVSAvoidrobustness to noise points
Core Design Contradiction:
Ease of operationVSReliability

Solution Approach 1:

The patent replaces the traditional Hough transform mechanical voting process with a direct polynomial fitting approach on connected component points. This substitution eliminates the sensitivity to noise points inherent in Hough transform while maintaining ease of operation, as the polynomial fitting on filtered points is computationally simple and directly yields lane line parameters.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP3617938B1Lane line processing method and device
Publication Date: 2024.02.28 APOLLO INTELLIGENT DRIVING (BEIJING) TECHNOLOGY CO LTD
  • EP3617938B1 patent drawingFigure 1
  • EP3617938B1 patent drawingFigure 2~3
  • EP3617938B1 patent drawingFigure 4~6

AI summary

Embodiments of the present application provide a lane line processing method and a lane line processing device. The method includes: performing a binarization processing on a first image to obtain a binary image, the first image including lane line points and non-lane line points; performing a connected domain analysis on the binary image to obtain at least one connected domain in the binary image, the connected domain including a plurality of adjacent lane line points; determining lane line points in a group corresponding to a lane line, based on the connected domain; and obtaining representation information of the lane line corresponding to the group, by using the lane line points in the group.