1A1-U10 音声と慣性センサに基づく災害救助犬の吠え検出

Abstract

This paper proposes audio-based and motion-based methods for detecting search and rescue (SAR) dogs' continuous barking. SAR dogs are trained to bark continuously when they find a victim. Therefore, the victim location can be detected based on using barking actions. Dynamic time warping (DTW) is used to analyze audio data. Periodic motion was happened during the dogs' barking action. The motion can be detected by an inertial measurement unit (IMU) on the vest. A fast Fourier transform (FFT) is used to analyze the dog's barking motion. The proposed detection methods were evaluated using audio and IMU data recorded during actual SAR dog trainings. The F-scores of the audio-based and motion-based barking detection methods were 0.95 and 0.90. As a trial, we marked victim locations on a map based on the body motion.