Neural Network Captioning for Unknown Traffic Signs

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current traffic sign recognition systems rely heavily on large databases and are unable to identify unknown signs, considering them as 'garbage' if not found within the database.

Innovation Solution

An advanced driving assistance system utilizing a combination of a traffic sign classifier and a neural network, specifically a CNN and LSTM network, to analyze images and generate captions for unknown traffic signs by training on a plurality of image data and associated captions for known signs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a large database of predefined traffic signs is used for recognition, then recognition accuracy for known signs is improved, but the system cannot identify unknown signs and considers them as garbage

Engineering Contradiction:
Improvetraffic sign recognition accuracyVSAvoidability to identify unknown signs
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent introduces an intermediary neural network component that bridges the gap between predefined sign recognition and unknown sign identification. The neural network acts as a mediator that receives images from the camera system and generates captions for unknown signs, enabling the system to handle signs not present in the predefined database while maintaining accurate recognition of known signs.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The system changes the parameter approach from static predefined sign categories to dynamic caption-based identification. Instead of matching images against fixed database entries, the system uses neural networks to generate descriptive captions that can adapt to any sign type, transforming the recognition paradigm from rigid classification to flexible description.

Inventive Principle:
Principle #35Parameter changes

2Device complexity

If a conventional traffic sign classification approach is used, then system simplicity is maintained, but the system cannot provide captions for unknown signs

Engineering Contradiction:
Improvesystem structure simplicityVSAvoidinformation about unknown signs
Core Design Contradiction:
Device complexityVSLoss of information

Solution Approach 1:

The patent segments the traffic sign recognition system into distinct functional components: a camera system for image acquisition, a classification system for known signs, and a neural network module for generating captions of unknown signs. This segmentation allows each component to specialize in specific tasks, maintaining overall system simplicity while enabling comprehensive sign identification and description.

Inventive Principle:
Principle #1Segmentation

3Ease of manufacture

If traffic signs not found in the database are considered garbage, then database maintenance simplicity is preserved, but useful information about unknown signs is lost

Engineering Contradiction:
Improvedatabase maintenance simplicityVSAvoidinformation about unknown traffic signs
Core Design Contradiction:
Ease of manufactureVSLoss of information

Solution Approach 1:

The neural network component enables the system to self-identify and self-describe unknown signs without requiring manual database updates or maintenance. The system automatically generates captions for unknown signs, effectively serving itself by acquiring and processing new sign information autonomously, eliminating the need for continuous human intervention in database maintenance.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP3557476B1Device for determining a caption for an unknown traffic sign
Publication Date: 2023.08.09 CONTINENTAL AUTONOMOUS MOBILITY GERMANY GMBH
  • EP3557476B1 patent drawingFigure 1~2
  • EP3557476B1 patent drawingFigure 3~4c
  • EP3557476B1 patent drawingFigure 5a~6

AI summary

The present invention relates to a device (10) for determining a caption for an unknown traffic sign. It is described to provide (310) a processing unit with at least one image data relating to an unknown traffic sign for which a caption associated with the traffic sign is not known. The processing unit implements (320) at least one artificial neural network to process the at least one image data to generate a caption for the unknown traffic sign. The at least one artificial neural network has been trained on the basis of a plurality of image data and associated captions for a plurality of known traffic signs, wherein the caption for each known traffic sign is known. An output unit outputs (330) the caption for the unknown traffic sign.