Training Data Correction for Road Surface Estimation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The estimation of road surface conditions using machine learning models is hindered by errors in manually set break times in training data, making high-accuracy learning difficult.

Innovation Solution

A training data generation device and method that corrects break times in time-series data by maximizing the difference between averages and variances of adjacent sections, using a formula to optimize the placement of labels and reduce errors.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If break times are manually set to divide sections in training data, then the training data can be constructed with labels assigned to each section, but errors of about a few milliseconds occur making highly accurate learning difficult

Engineering Contradiction:

Improvebreak time accuracyVSAvoidlabel assignment accuracy

Core Design Contradiction:

Measurement precisionVSManufacturing precision

Solution Approach 1:

The system automatically corrects break times using the sensor data itself rather than relying on manual setting. The correction unit uses the actual sensor data characteristics (variance, average values) to self-determine the optimal break time positions, eliminating the need for manual intervention and achieving both operational simplicity and high precision simultaneously

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The manual mechanical process of setting break times is replaced with an automated computational system. The correction unit uses mathematical calculations based on sensor data statistics (variance ratios, average differences) to automatically determine and correct break time positions, substituting human manual operation with a computational mechanism that achieves higher precision

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Ease of manufacture

If manual section division is used to assign labels, then the process is simple to implement, but errors occur that prevent high-accuracy learning

Engineering Contradiction:

Improvetraining data construction easeVSAvoidlearning accuracy

Core Design Contradiction:

Ease of manufactureVSReliability

Solution Approach 1:

The system performs preliminary correction of break times before the actual learning process. By automatically adjusting break time positions based on sensor data characteristics in advance, the system ensures that the training data is optimally prepared, eliminating the need for complex manual adjustments while ensuring high learning accuracy

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The correction unit uses feedback from the sensor data itself to automatically adjust break time positions. By calculating variance ratios and average differences from the actual sensor data and using this feedback to determine optimal break points, the system achieves accurate label assignment without manual intervention, combining ease of implementation with high reliability

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS12118438B2Learning data generation device, learning data generation method, and program

Publication Date: 2024.10.15 NIPPON TELEGRAPH & TELEPHONE CORP

AI summary

A training data generation device (10) according to the present invention includes a training data correction unit (11) configured to correct training data that is time-series data that indicates states of an object and in which a label corresponding to a state of the object is assigned to each section that corresponds to the state indicated by the label, wherein the training data correction unit (11) corrects a break time that divides a first section from a second section adjacent to the first section, based on time-series data in the first section and time-series data in the second section.