Real-time audio monitoring and voice communication
Rescue teams can listen to sounds from the entrapment site via a headset and communicate directly with survivors near the sensors.
Search and Rescue Survivors
Supports survivor search and rescue operations.
Intelligent wireless acoustic sensors and AI support survivor search and rescue operations at disaster sites.
Rescue clues generated by sensors and AI
Rescuers can locate and rescue trapped victims based on position information identified through multi-sensor acoustics and AI analysis.
Real-time victim localization through AI analysis
AI analyzes and integrates distributed sensor signals so operators can monitor locations in real time.
Intelligent acoustic sensors deployed on site
Sensors deployed at the incident site collect sounds and vibrations generated beneath debris.
Equipment Overview Video
Demonstrates how to use the acoustic sensors and the terminal device through real on-site scenarios.
AI Acoustic Detection & Localization
Abnormal sound and voice detection and localization technology
AI-based acoustic analysis and sensor data detect survivor signals and estimate their location.
AI Acoustic Detection
Detects acoustic signals of survivors even in disaster environments
Even in complex noisy environments, AI can identify subtle acoustic signals from survivors.
FiNDU AI utilizes a Transformer-based Acoustic Spectrogram Transformer (AST) model trained on various real disaster-environment audio datasets to detect abnormal sounds from survivors. Through multi-label acoustic classification, the system analyzes survivor sounds and environmental noise simultaneously, enabling stable detection even in high-noise environments.
AI-Based Localization
Estimates survivor locations through multi-sensor analysis
Signals collected from multiple sensors become key clues for estimating the location of survivors.
FiNDU AI uses TDoA (Time Difference of Arrival) based localization technology to analyze time differences in signals collected from multiple acoustic sensors. By integrating sensor data, the system estimates survivor locations and helps narrow the search area effectively in rescue operations.
Intelligent USAR System Architecture
Sound Source Localization Modeling
We are validating and improving the localization model through ongoing field verification tests.
Modeling and Experiments
- Experiment: Sound source localization using simulation and field-verification datasets
- Experiment: Data is downsampled to 8,000 Hz
- Step 1.Separate signals using a filter bank
- Step 2.Estimate the GCC-PHAT PDF using Kernel Density Estimation (KDE)
- Step 3.Generate hyperbolic curves based on TDOA
- Step 4.Estimate the optimal coordinates
Noise Reduction and Sound Recognition Modeling
Core Equipment Introduction
