Automated Lane Marker Labeling via Map Projection Alignment

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Reliable lane detection is crucial for driver assistance and fully automated vehicles, but existing technologies face challenges in accurately detecting lane markers without vast amounts of labeled data, especially in varying environmental conditions.

Innovation Solution

An automated system that projects high-definition maps onto camera images, aligns features, and generates labels for lane markers using an electronic processor, enabling efficient training of neural networks for lane detection without manual labeling, even in grayscale mono camera inputs, and allowing detection of lane markers up to 150 meters away.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If deep learning is used for lane marker detection, then detection accuracy is improved, but the need for vast amounts of manually labeled data increases

Engineering Contradiction:
Improvelane marker detection accuracyVSAvoidmanual data labeling time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary actions by using high-definition map data to pre-generate labels for lane markers before the deep learning training process. This preliminary labeling based on authoritative map data eliminates the need for manual annotation, providing ready-to-use training data that enables accurate lane detection without time-consuming manual work

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system creates copies of lane marker information from high-definition maps and projects them onto camera images. These copied and projected map features serve as synthetic labels that replicate real lane marker appearances and positions, providing abundant training data without requiring manual capture or annotation of actual road scenes

Inventive Principle:
Principle #26Copying

2Measurement precision

If high-definition maps are projected onto camera images, then label generation accuracy is improved, but misalignments due to localization inaccuracies occur

Engineering Contradiction:
Improvelabel generation accuracyVSAvoidprojection alignment precision
Core Design Contradiction:
Measurement precisionVSManufacturing precision

Solution Approach 1:

The system implements feedback by detecting actual lane markers in camera images and comparing their positions with projected map features. The detected positions provide feedback that is used to refine and correct the projection transformation, iteratively improving alignment accuracy and compensating for initial localization errors

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The system replaces mechanical alignment methods with computational image processing techniques. Instead of relying solely on physical localization accuracy, the system uses digital image analysis to detect lane markers and computationally adjusts the projection parameters to achieve precise alignment between map features and actual image content

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

Data Source

PatentEP3607272B1Automated image labeling for vehicle based on maps
Publication Date: 2022.04.27 ROBERT BOSCH GMBH
  • EP3607272B1 patent drawingFigure 1~2
  • EP3607272B1 patent drawingFigure 3
  • EP3607272B1 patent drawingFigure 4~5

AI summary

A method and system for navigating a vehicle using automatically labeled images. The system includes a camera configured to capture an image of a roadway and an electronic processor communicatively connected to the camera. The electronic processor is configured to load a map that includes a first plurality of features and receive the image. The image includes a second plurality of features. The electronic processor is further configured to project the map onto the image; detect the second plurality of features within the image; and align the map with the image by aligning the first plurality of features with the second plurality of features. The electronic processor is further configured to copy a label describing one of the first plurality of features onto a corresponding one of the second plurality of features to create a labelled image and to use the labelled image to assist in navigation of the vehicle.