End-to-End Lane Line Recognition With Basic and Local Feature Fusion

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing lane line detection methods based on deep learning suffer from poor recognition accuracy due to redundant post-processing steps, leading to cumulative recognition errors.

Innovation Solution

A lane line recognition method that integrates end-to-end processing by extracting basic and local features, fusing them, and using a novel coding scheme to represent lane line information as a three-dimensional vector for improved accuracy and robustness.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If semantic segmentation and post-processing lane line extraction are used, then lane line detection can be performed, but recognition accuracy deteriorates due to cumulative errors from redundant steps

Engineering Contradiction:
Improvelane line recognition accuracyVSAvoidnumber of processing steps
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts and removes the redundant post-processing steps from the traditional semantic segmentation pipeline. By directly outputting lane line information from the neural network without intermediate processing steps, it eliminates the source of cumulative errors while maintaining detection functionality.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent merges the lane line detection and recognition functions into a single integrated neural network output. Instead of separate processing stages for segmentation followed by extraction, the model directly produces lane line parameters, combining multiple functions into one unified operation that reduces error accumulation.

Inventive Principle:
Principle #5Merging (Combining)

2Reliability

If multiple post-processing steps are applied, then lane line extraction can be performed, but recognition reliability worsens due to error accumulation

Engineering Contradiction:
Improvelane line recognition reliabilityVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent extracts and eliminates the problematic post-processing steps that cause error accumulation. By removing these redundant operations, the system maintains reliability while actually reducing processing time, as fewer steps mean fewer opportunities for errors to accumulate.

Inventive Principle:
Principle #2Taking out (Extraction)

3Manufacturing precision

If traditional post-processing extraction is used, then lane line information can be obtained, but manufacturing precision (recognition accuracy) deteriorates

Engineering Contradiction:
Improvelane line recognition precisionVSAvoidsystem implementation simplicity
Core Design Contradiction:
Manufacturing precisionVSEase of manufacture

Solution Approach 1:

The patent extracts the essential lane line detection capability from the complex post-processing pipeline and integrates it directly into the neural network output. This simplifies the system implementation while improving precision, as the direct output approach eliminates the need for multiple processing stages that reduce accuracy.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP4332912B1Lane line recognition method and apparatus, electronic device and storage medium
Publication Date: 2025.10.15 BEIJING BAIDU NETCOM SCI & TECH CO LTD
  • EP4332912B1 patent drawingFigure 1~2
  • EP4332912B1 patent drawingFigure 3~4
  • EP4332912B1 patent drawingFigure 5

AI summary

The present disclosure provides a lane line recognition method and apparatus (700, 800), an electronic device and a storage medium, relates to a technical field of artificial intelligence, in particular to technical fields of intelligent transportation, automatic driving and deep learning. The specific implementation includes: extracting (S210) a basic feature of an original image; recognizing (S220) at least one lane line node in the original image by using the basic feature of the original image; extracting (S230) a local feature from the basic feature of the original image by using the at least one lane line node, and fusing the basic feature and the local feature; and recognizing (S240) a lane line in the original image based on a fused result. The present disclosure can improve lane line recognition effect.