Online multi-target acoustic tracking for mobile assets

Passive acoustic sensing is increasingly used for environmental and situational awareness in autonomous maritime systems. However, real-time processing on resource-constrained edge devices poses significant challenges. To address this, we propose a lightweight, automated multi-target tracking system optimized for low latency (∼1.7 ms) and low power (∼1.3 W) on a Raspberry Pi Zero 2W. Our method employs a hybrid cluster-filter approach, combining DBSCAN clustering for estimating the number of sources and Kalman filtering for continuous tracking of targets in terms of their direction rather than absolute position. We demonstrate the system’s effectiveness using passive acoustic recordings of diving beaked whales in open ocean conditions. Neural networks with post-training optimization are integrated to enhance performance while minimizing computational demands. Designed for cross-platform flexibility, this system can operate on mobile assets like gliders, drifters, and profilers. Preliminary results highlight its potential for real-time autonomous underwater applications, offering a robust, energy-efficient alternative to existing multi-target tracking implementations.