Vehicle Visual Object Classification With Uncertainty-Based Active Learning

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The high cost and inefficiency of generating large, labeled datasets for training machine learning algorithms in vehicles for partial driving automation, and the need for accurate active learning criteria to enhance object classification in automotive perception tasks.

Innovation Solution

An active learning method that determines uncertainty values and patch distance measures for automotive sensor data to selectively request annotations from an oracle, ensuring accurate and efficient training of object classifiers in vehicles.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If large, labeled datasets are generated manually to train machine learning algorithms for automotive perception tasks, then the accuracy and reliability of object classification improve, but the cost and time required for data preparation increase significantly

Engineering Contradiction:
Improveaccuracy of object classificationVSAvoidtime for data preparation
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system performs self-training by automatically identifying uncertain predictions from the machine learning model and selecting these samples for manual annotation. This self-service mechanism allows the system to autonomously determine which data points need human labeling, significantly reducing the time and cost of data preparation while maintaining high accuracy through targeted annotation of only the most uncertain cases

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

Instead of manually labeling entire datasets, the system applies partial action by selectively annotating only a subset of data points that have high uncertainty values. This partial annotation approach achieves sufficient training accuracy without the excessive time investment required for complete dataset labeling, resolving the contradiction between accuracy and time consumption

Inventive Principle:
Principle #16Partial or excessive action

2Productivity

If active learning is used to reduce manual labeling effort, then the cost and time of data preparation decrease, but the accuracy of the machine learning algorithm may be insufficient for at least partial driving automation

Engineering Contradiction:
Improveefficiency of data preparationVSAvoidaccuracy of machine learning algorithm
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The system changes the parameter selection criterion from random or uniform sampling to uncertainty-based sampling. By using uncertainty values as the selection parameter, the system identifies and annotates only the most informative data points that will most improve model accuracy. This parameter change enables high productivity through selective annotation while ensuring the machine learning algorithm achieves the required accuracy level for driving automation

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The system implements a feedback loop where the machine learning model's uncertain predictions are fed back to identify samples for manual annotation. The annotated samples are then used to retrain and improve the model, creating a continuous cycle of improvement. This feedback mechanism ensures that each round of partial annotation systematically reduces uncertainty and improves accuracy, resolving the contradiction between productivity and reliability

Inventive Principle:
Principle #23Feedback

3Productivity

If uncertainty-based active learning is applied to select data points for annotation, then the cost of data preparation is reduced, but additional complexity is introduced in determining uncertainty values and selecting patches

Engineering Contradiction:
Improveefficiency of data preparationVSAvoidcomplexity of active learning criteria
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The system segments the data selection process into distinct components: uncertainty calculation for individual data points, patch formation by grouping spatially related uncertain points, and patch distance calculation for diversity selection. This segmentation of the complex active learning process into manageable modules reduces the perceived complexity while maintaining high productivity through systematic, step-by-step data selection

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4604079A1Uncertainty based active learning for object classification in visual perception tasks in a vehicle
Publication Date: 2025.08.20 BAYERISCHE MOTOREN WERKE AG
  • EP4604079A1 patent drawingFigure 1
  • EP4604079A1 patent drawingFigure 2A
  • EP4604079A1 patent drawingFigure 2B

AI summary

The present invention relates to enabling active learning for object classification in visual perception tasks in a vehicle. To this end, an object class out of a plurality of object classes is determined for one or more data points within automotive sensor data. Further, an uncertainty value for each of the one or more data points within the automotive sensor data indicative of an uncertainty of the object class determination and one or more object instances within the automotive sensor data are determined. Then, a patch for each data point having a corresponding uncertainty value exceeding an uncertainty threshold and a patch distance measure for each patch is determined. Each patch having a patch distance measure exceeding a patch distance measure threshold is provided to an oracle. Finally, an object annotation of each patch is received from the oracle.