3D Traffic Management Object Annotation Using Multi-Frame 2D Boxes

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing 2D annotation methods for traffic management objects in autonomous driving systems are inadequate for capturing depth, orientation, and spatial relationships, and manual 3D annotation is labor-intensive, time-consuming, and prone to errors.

Innovation Solution

A method for automatic 3D annotation using 2D bounding boxes in multiple image frames to localize 3D bounding boxes, generate 2D image cutouts, and train models to classify traffic management objects, leveraging neural networks and geometric calculations to enhance accuracy and efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual 3D annotation is performed, then annotation accuracy is improved, but productivity deteriorates due to labor-intensive and time-consuming processes

Engineering Contradiction:
Improveannotation accuracyVSAvoidannotation speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent replaces manual mechanical annotation processes with an automated computer vision system that uses neural networks and geometric calculations to perform 3D annotation, thereby maintaining accuracy while dramatically improving productivity

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

Solution Approach 2:

The system uses 2D bounding boxes from existing images as copies or projections to generate 3D annotations automatically, eliminating the need for manual 3D marking while preserving annotation quality through mathematical transformation

Inventive Principle:
Principle #26Copying

2Productivity

If 2D annotation is used, then productivity is improved, but measurement precision deteriorates due to inability to capture depth and spatial relationships

Engineering Contradiction:
Improveannotation efficiencyVSAvoidspatial relationship accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent transforms 2D annotation data into 3D spatial information by introducing depth dimension through geometric calculations and neural network predictions, thereby maintaining efficiency while gaining accurate spatial relationships

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

3Manufacturing precision

If manual 3D annotation is performed, then annotation detail is improved, but loss of time increases due to the complex and slow process

Engineering Contradiction:
Improveannotation detailVSAvoidannotation time
Core Design Contradiction:
Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent substitutes manual annotation mechanics with automated computational processes that use neural networks and geometric transformations to generate detailed 3D annotations instantly, eliminating time loss while preserving detail quality

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

4Productivity

If automated 3D annotation methods are used, then productivity is improved, but manufacturing precision may deteriorate due to algorithmic approximations

Engineering Contradiction:
Improveannotation throughputVSAvoidannotation accuracy
Core Design Contradiction:
ProductivityVSManufacturing precision

Solution Approach 1:

The patent uses neural networks trained on labeled data to perform automated 3D annotation with high precision, replacing manual methods while maintaining or exceeding accuracy through learned patterns and geometric constraints

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

Solution Approach 2:

The system incorporates feedback mechanisms where annotation results are validated and refined through iterative processing, ensuring that automated methods maintain high precision by correcting algorithmic approximations

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP4645249A1Three-dimensional annotation of traffic management objects
Publication Date: 2025.11.05 AIMOTIVE KFT
  • EP4645249A1 patent drawingFigure 1
  • EP4645249A1 patent drawingFigure 2
  • EP4645249A1 patent drawingFigure 3

AI summary

The disclosure refers to a computer-implemented method for annotating traffic management objects in three-dimensional (3D) space. The method includes obtaining a plurality of image frames, at least some of the image frames including two-dimensional (2D) bounding boxes of traffic management objects, and each image frame associated with position and orientation data, localizing centers of 3D bounding boxes of traffic management objects using the plurality of image frames including the 2D bounding boxes, and the respective position and orientation data, for each localized center of a 3D bounding box of a traffic management object, calculating an extent and orientation of the 3D bounding box based on image frames including the 2D bounding box of the traffic management object, for each 3D bounding box of a traffic management object, projecting the 3D bounding box on image frames associated with the traffic management object, generating a plurality of 2D image cutouts of traffic management objects based on the projected 3D bounding boxes, and training at least one model to classify traffic management objects in the plurality of 2D image cutouts. Furthermore, a method of generating a training dataset, and corresponding devices are disclosed.