Lane Identification Using Multi-Virtual Lanes and Semantic Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current lane departure warning systems and adaptive cruise control technologies are limited in processing an entire road structure and recognizing multiple lanes, as they typically rely on narrow viewing angles and do not account for occlusions or entire road environments, leading to incomplete lane identification and vehicle positioning.

Innovation Solution

A method and apparatus that utilize convolutional neural networks and semantic segmentation to extract road markings and generate multi-virtual lanes from input images, allowing for the identification of the driving lane by analyzing the relative location within a segmented and transformed image, enabling the detection of lane boundaries and signs across the entire road.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If a narrow viewing angle is used for lane detection, then the system complexity is reduced, but the ability to recognize entire road structure and multiple lanes deteriorates

Engineering Contradiction:
Improvesystem complexityVSAvoidability to recognize entire road structure
Core Design Contradiction:
Device complexityVSAdaptability or versatility

Solution Approach 1:

The patent transforms the 2D input image into a top-view image through perspective transformation, adding a dimensional perspective change that allows the system to recognize the entire road structure and multiple lanes while maintaining manageable system complexity. This dimensional transformation enables comprehensive road awareness without requiring physically complex multi-camera setups.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Adaptability or versatility

If the entire road is captured in the input image, then the ability to identify multiple lanes is improved, but the difficulty of detecting and measuring road markings increases

Engineering Contradiction:
Improveidentification of multiple lanesVSAvoiddetection of road markings
Core Design Contradiction:
Adaptability or versatilityVSDifficulty of detecting and measuring

Solution Approach 1:

The patent segments the road detection task into distinct processing stages: first performing perspective transformation to create a top-view image, then detecting road markings in the transformed space, and finally mapping results back to the original coordinate system. This segmentation of the detection process makes it easier to identify multiple lanes while managing the complexity of detecting road markings across the entire road.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces a top-view image as an intermediary representation between the original input image and the final lane identification results. This intermediary transformation simplifies the detection of road markings by presenting them in a geometrically corrected view where perspective distortions are eliminated, making multiple lane detection more reliable.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Measurement precision

If perspective transformation is applied to generate top-view image, then the accuracy of lane identification is improved, but the processing time increases

Engineering Contradiction:
Improveaccuracy of lane identificationVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs perspective transformation as a preliminary step before road marking detection, pre-processing the image into a top-view representation that simplifies subsequent detection tasks. By preparing the image in advance with the correct geometric perspective, the actual lane identification process becomes faster and more accurate, offsetting the initial transformation time.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3462377B1Method and apparatus for identifying driving lane
Publication Date: 2024.02.14 SAMSUNG ELECTRONICS CO LTD
  • EP3462377B1 patent drawingFigure 1
  • EP3462377B1 patent drawingFigure 2
  • EP3462377B1 patent drawingFigure 3

AI summary

A method and apparatus for identifying a driving lane extracts a road marking from an input image, generate a multi-virtual lane using the road marking and a segmentation image obtained by segmenting the input image into objects included in the input image based on a semantic unit, and identifies a driving lane of a vehicle based on a relative location of the driving lane in the multi-virtual lane.