Behavior Identification Using Spectrum Pattern Extraction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional behavior identification devices struggle to accurately identify target behaviors when surrounding sounds include noise or other non-target behaviors, leading to reduced accuracy and malfunction.

Innovation Solution

The method involves acquiring surrounding sound, extracting feature values from spectrum patterns stored in a storage unit, and using these feature values to identify the target behavior through a machine-learned model, even in the presence of noise, by focusing on the frequency patterns and their levels in the sound data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional behavior identification devices use sound-based recognition, then they can identify target behaviors, but they malfunction when other behaviors or noise are present

Engineering Contradiction:

Improvebehavior identification accuracyVSAvoidsystem stability in noisy environments

Core Design Contradiction:

Measurement precisionVSReliability

Solution Approach 1:

The patent extracts and isolates specific spectrum patterns characteristic of target behaviors from the overall sound signal. By separating and focusing on these distinctive spectral features, the system can identify target behaviors even when other sounds or noise are present in the environment.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent introduces spectrum information as an intermediary representation between the raw sound signal and behavior identification. This spectral domain transformation serves as a mediator that enhances the distinguishability of target behavior sounds from background noise and other behaviors.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Measurement precision

If the system focuses on identifying target behavior sounds, then identification accuracy improves, but the system becomes vulnerable to false identification from similar sounds

Engineering Contradiction:

Improvetarget behavior identification accuracyVSAvoidfalse identification from similar sounds

Core Design Contradiction:

Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent applies local quality analysis by examining specific local characteristics of the spectrum pattern rather than overall sound properties. By focusing on distinctive spectral features at specific frequencies and time segments, the system can accurately identify target behaviors while distinguishing them from similar but different sounds.

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS11501209B2Behavior identification method, behavior identification device, non-transitory computer-readable recording medium recording therein behavior identification program, machine learning method, machine learning device, and non-transitory computer-readable recording medium recording therein machine learning program

Publication Date: 2022.11.15 PANASONIC INTELLECTUAL PROPERTY CORP OF AMERICA

AI summary

In a behavior identification method, surrounding sound is acquired, a feature value that is specified by a spectrum pattern included in spectrum information generated from sound made by a person performing a predetermined behavior is extracted from the sound acquired, the predetermined behavior is identified by the feature value, and information indicating the predetermined behavior identified is output.