Vehicle Classifier Updating Using Backend Consistency Voting

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Manual identification and annotation of training samples for vehicle-side classifiers in advanced driver assistance systems is time-consuming and costly, and existing methods for updating classifiers are inefficient and resource-intensive.

Innovation Solution

A backend-side classifier is trained using classification data from vehicle-side classifiers, with a consistency check to identify and correct inconsistent results, and a voting scheme to determine consistent annotations, allowing for efficient and automated updating of vehicle-side classifiers.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual identification and annotation of training samples is used, then classification accuracy can be improved, but time consumption and cost increase significantly

Engineering Contradiction:
Improveclassification accuracyVSAvoidtime consumption
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system uses vehicle-side classifiers to automatically generate classification results that serve as annotations for training backend-side classifiers. The classifiers self-upgrade by using their own output to train improved versions, eliminating the need for manual annotation while maintaining accuracy improvement.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The backend receives classification results from vehicles, applies consistency checks to identify correct annotations, and uses these feedback annotations to retrain and improve the backend classifier. This closed-loop feedback mechanism continuously improves accuracy without manual intervention.

Inventive Principle:
Principle #23Feedback

2Reliability

If manual annotation of training samples is performed, then reliable training data can be obtained, but cost increases significantly

Engineering Contradiction:
Improvetraining data reliabilityVSAvoidannotation cost
Core Design Contradiction:
ReliabilityVSEase of manufacture

Solution Approach 1:

The system generates its own training annotations through vehicle-side classifiers rather than requiring external manual annotation services. The classifiers produce their own labeled training data, making the process self-sufficient and eliminating annotation costs.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The backend-side classifier acts as an intermediary that receives raw classification results from vehicles, applies consistency checks to determine reliability, and converts these into reliable training annotations. This intermediary process ensures reliability without direct manual intervention.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Measurement precision

If classification results from multiple vehicles are collected for consistency check, then annotation accuracy can be improved, but communication resources are consumed

Engineering Contradiction:
Improveannotation accuracyVSAvoidcommunication resources
Core Design Contradiction:
Measurement precisionVSLoss of energy

Solution Approach 1:

The system collects classification results from multiple vehicles but only processes a subset through consistency checks. By applying consistency checks selectively rather than universally, the system achieves sufficient annotation accuracy while limiting communication resource consumption to necessary levels.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3724808B1Method and system for training and updating a classifier
Publication Date: 2026.03.04 CONTINENTAL AUTOMOTIVE TECHNOLOGIES GMBH
  • EP3724808B1 patent drawingFigure 1
  • EP3724808B1 patent drawingFigure 2
  • EP3724808B1 patent drawingFigure 3

AI summary

The present disclosure relates to a method for updating and training a backend-side classifier. The method includes receiving, in a backend-device, from at least one vehicle classification data along with a respective classification result generated by a vehicle-side classifier and training the backend-side classifier using the classification data and a possibly corrected respective classification result as annotation. Accordingly, the backend-side classifier is trained or fine-tuned based on the result of a pre-trained vehicle-side classifier. Classification results and the underlying classification data are sent to the backend by one or more vehicles.