Autonomous Splash Detection System for Data Collection: A Subaqueous Multi-DAS Beamforming Approach

Grevenkop-Castenskiold, Bjørn Højmose ; Nielsen, Jonas Emil

4. semester

Electronic Systems, Master

2024

2024-05-31

128

Formålet med projektet er at opbygge et system, der autonomt kan opdage, optage og uploade plask fra mennesker ved badeområder for at skabe et datasæt til træning af algoritmer, der kan opdage potentielle drukneulykker i havne. Systemet arbejder i flere trin: først opdages et plask, derefter beregnes lydens retning (Direction of Arrival, DOA) ud fra små forskelle i ankomsttid mellem mikrofonerne (Time Difference of Arrival, TDOA). Systemet bruger delay‑and‑sum (DAS) beamforming til at fokusere lytningen mod den formodede retning, bekræfter hændelsen og optager og uploader data til en server via Wi‑Fi. En undervands DAS‑beamformer blev implementeret på en ESP32‑S3‑mikrocontroller. Den består af tre uniforme lineære arrays (ULAs), hver målrettet mod sin egen frekvensbånd for at opretholde retningsfølsomhed og undertrykkelse af uønskede retninger på tværs af det samlede frekvensområde. Systemet modtager lyd fra syv 24‑bit I2S‑mikrofoner. Mikrofonerne blev akustisk koblet til vandet (impedanstilpasset) med nitrilhandsker for bedre lydoverførsel. Lyden sendes over SPI fra en PSoC 5LP til ESP32‑S3, hvorefter data offloades til en server. Et automatisk testsystem blev brugt til at verificere beamformerens "beampattern" (følsomhed mod forskellige retninger) i luft. Testene bestod i at afspille og optage en eksponentiel sinus‑sweep (ESS) og enkelttoner (STT) for hver vinkel i et anekoisk rum (et rum, der dæmper ekko). De målte beampatterns blev sammenlignet med simuleringer, hvor der blev fundet god overensstemmelse samt identificeret nogle støjkilder. Systemet er endnu ikke evalueret i vand, men vurderes klar til udplacering med enkelte yderligere udviklinger og nødvendige tilladelser.

This project builds a system that autonomously detects, records, and uploads splash events at bathing locations to create a dataset for training algorithms that can reliably spot potential drowning incidents in harbors. The system operates in stages: it detects a splash, estimates the sound’s direction (Direction of Arrival, DOA) using small differences in arrival time between microphones (Time Difference of Arrival, TDOA), applies delay‑and‑sum (DAS) beamforming to focus listening toward that direction, confirms the event, and records and uploads data to a server via Wi‑Fi. An underwater DAS beamformer was implemented on an ESP32‑S3 microcontroller. It uses three uniform linear arrays (ULAs), each tuned to its own frequency band to maintain directional sensitivity and suppression of unwanted directions across the full range. Audio is captured by seven 24‑bit I2S microphones. The microphones were acoustically coupled to water (impedance matched) using nitrile gloves to improve sound transfer. Audio is forwarded over SPI from a PSoC 5LP to the ESP32‑S3, and data are offloaded to a server. An automatic test system verified the beamformer’s beampattern (sensitivity to sounds from different directions) in air. Tests played and recorded an exponential sine sweep (ESS) and single tones (STT) at each angle inside an anechoic chamber (a room designed to absorb reflections). Measured beampatterns were compared to simulations, showing good similarity along with some noise sources. The system has not yet been evaluated in water, but is considered close to deployment with minor further development and required permits.

[This abstract was generated with the help of AI]

Beamforming ; Hydrophone ; DAS ; Differential ; ESP32 ; PSoC ; anechoic chamber ; drowning ; harbour ; safety ; detection ; autonomous ; dataset ; training data

Download
View record in AAU Student Projects