Highlights

[IoT 2024] Sailing Through Data: Automated Data Acquisition Systems for Machine Learning Applications in Maritime Navigation

We have introduced AISail, a custom embedded hardware and software platform designed to automate data acquisition for machine learning in maritime navigation. This work, “Sailing Through Data: Automated Data Acquisition Systems for Machine Learning Applications in Maritime Navigation,” was peer-reviewed and presented at the 14th International Conference on the Internet of Things (IoT 2024), a recognized international venue in IoT research.

AISail integrates RGB and infrared cameras, GPS, an inertial measurement unit, and an AIS antenna into a portable, self-contained system capable of real-time deployment onboard ships or at coastal stations. Its key innovation is an automated data annotation pipeline that fuses ship detections from computer vision models with Automatic Identification System (AIS) data, producing richly annotated datasets that include bounding boxes, ship category, distance, speed, and unique identifiers. To further improve accuracy, the system incorporates an orientation correction module based on simulated annealing, aligning AIS signals with camera detections more precisely.

Unlike existing maritime datasets, which are often based on static webcams or third-party AIS services, AISail generates paired RGB and IR images with real-world distance and speed annotations, a feature not available elsewhere. This capability enables the creation of high-quality datasets tailored for machine learning applications in ship detection, tracking, and navigation.

The applications of AISail are wide-ranging. In maritime navigation and collision avoidance, it enhances situational awareness for vessels. For port and coastal surveillance, it supports monitoring of illegal fishing, smuggling, and pollutant dumping. In fisheries and environmental monitoring, it can track small vessels not broadcasting AIS signals. Beyond operational use, AISail serves as a powerful tool for dataset generation, supporting the training and validation of detection and tracking models. Its relevance extends to defense and security domains, where monitoring non-cooperative or small vessels is critical, and to the future of autonomous ships and maritime robotics, where reliable ship detection and distance estimation are essential.

AISail is more than just a system. It is a gateway to new research. By providing richly annotated maritime data, it enables future researchers to develop and train advanced detection and distance estimation models without the heavy burden of building a dataset from scratch.