Lane Marker Detection via End-to-End CNN and Horizontal Reduction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing lane detection systems in vehicles are slow, require significant manual configuration, and lack robustness across various driving scenarios, making them unsuitable for modern vehicles with advanced driver assistance systems.

Innovation Solution

An efficient end-to-end lane marker detection architecture that directly predicts lane marker vertices using a convolutional neural network, reducing computational complexity by modeling lane markers as a row-wise classification task and utilizing horizontal reduction modules to compress spatial information, without costly post-processing steps.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If existing lane detection techniques are used, then lane detection can be performed, but the system is slow and requires significant manual configuration

Engineering Contradiction:
Improvelane detection speedVSAvoidmanual configuration requirements
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The system performs self-configuration by automatically adapting to different driving scenarios and lane marker types without requiring manual setup. The neural network model learns optimal detection parameters during training and adjusts automatically during operation, eliminating the need for manual configuration while maintaining high detection speed.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system dynamically adjusts detection parameters based on input image characteristics and driving conditions. By changing parameters such as detection thresholds, region of interest boundaries, and model configuration automatically, the system achieves both high speed and adaptability without manual intervention.

Inventive Principle:
Principle #35Parameter changes

2Reliability

If existing lane detection techniques are used, then lane detection can be performed, but the system lacks robustness across many driving scenarios

Engineering Contradiction:
Improverobustness across driving scenariosVSAvoiddetection speed
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The detection system is divided into specialized components including shared backbone features, lane-specific heads for different lane marker types (solid, dashed, curved), and scenario-specific processing modules. This segmentation allows each component to be optimized for specific tasks while maintaining overall system robustness and efficient parallel processing.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system employs a universal neural network architecture that can detect multiple types of lane markers (solid lines, dashed lines, curved lines, construction zones) using a single trained model. This multi-functional approach ensures robustness across diverse driving scenarios while maintaining fast detection speed through unified processing.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Measurement precision

If traditional lane detection methods with post-processing are used, then accurate lane markers can be detected, but computational complexity and time increase

Engineering Contradiction:
Improvelane marker detection accuracyVSAvoidcomputational time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system extracts and removes the computationally expensive post-processing steps from the detection pipeline. By directly outputting final lane marker parameters from the neural network without requiring subsequent geometric processing or validation steps, the system maintains high accuracy while significantly reducing computational time.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

Traditional mechanical post-processing algorithms are replaced with a learned neural network model that directly predicts lane marker parameters. This substitution transfers the detection logic from explicit algorithmic processing to implicit learned patterns, achieving both high accuracy and speed through end-to-end learning.

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

Data Source

PatentEP4118568B1Lane marker detection
Publication Date: 2026.02.25 QUALCOMM INC
  • EP4118568B1 patent drawingFigure 1
  • EP4118568B1 patent drawingFigure 2
  • EP4118568B1 patent drawingFigure 3

AI summary

Certain aspects of the present disclosure provide a method for lane marker detection, including: receiving an input image; providing the input image to a lane marker detection model; processing the input image with a shared lane marker portion of the lane marker detection model; processing output of the shared lane marker portion of the lane marker detection model with a plurality of lane marker-specific representation layers of the lane marker detection model to generate a plurality of lane marker representations; and outputting a plurality of lane markers based on the plurality of lane marker representations.