← Back to Sessions

Development of a Cots-Based Platform for Real-Time Seismic and Acoustic Vehicle Detection and Characterization

Description: 

We are developing a system for real-time seismic and acoustic detection and characterization of vehicles utilizing Artificial Intelligence at the computing Edge (Edge AI). Seismic and acoustic sensors deployed in urban and industrial environments can capture information related to pattern of life as well as operational events (e.g., machinery operation, vehicle traffic, and environmental control units). In particular, signals from approaching vehicles can be observed at very local distances (few tens of meters) as short pulses of energy (lasting a few seconds) with spectral content from a few 10s of Hz to a few kHz. Given the broad bandwidth features of these signals and the potential of high number of events, processing at the sensing point is preferable to storing or transmitting full data for post-processing. Furthermore, processing at the collection point can exploit multi-channel data analysis (e.g., array processing or polarization) for movement tracking. Our system utilizes an inexpensive Commercial-of-the-Shelf Linux-box computer with an acquisition card to collect multichannel seismic and acoustic waveform data and apply tensorflow (TF) lite models for vehicle detection and type identification. The workflow for the unit is based on running an anomaly detection (i.e., short-term-average-over-long-term-average) algorithm to capture the pulses and apply TF for the characterization. We are collecting data at an industrial environment for the development of the TF models focused on utility vehicles. As running TF required significant computational resources, we are testing the performance of the hardware and software under different computational loads (e.g.: number of channels and sampling rate) for the operation of the full workflow of the system: multichannel acquisition, detection, and characterization.

Session: Detecting, Locating, Characterizing and Monitoring Non-earthquake Seismoacoustic Sources

Type: Oral

Date: 4/19/2023

Presentation Time: 03:00 PM (local time)

Presenting Author: Omar Marcillo

Student Presenter: No

Invited Presentation: 

Authors