Work

Method to Detect Only Registered Pedestrian Using Out‐of‐Distribution Detection and CNN Classifier with Footsteps

Year: 2025

Type: article

Abstract: A method for detecting registered pedestrians using footstep sounds is investigated. Pre‐trained persons were classified based on out‐of‐distribution detection and subsequently identified using a convolutional neural network classifier. As a training feature, the log mel‐spectrogram is calculated by a time–frequency analysis of the footstep sounds. The evaluation test results with newly prepared footsteps indicated the feasibility of the proposed method. © 2025 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.

Source: IEEJ Transactions on Electrical and Electronic Engineering

DOI: 10.1002/tee.70036

Authors: Hidefumi Kamozawa, Takayoshi Yamahira, Motoshi Tanaka

Institution: Akita University

Language: English

Topic: Music and Audio Processing

Subfield: Signal Processing

Field: Computer Science

Domain: Physical Sciences

Open Access status: bronze

Funder: Japan Society for the Promotion of Science

Award: 因果推論を用いた独居者生活活動音の確率モデル最適化と異常検出方法の開発