Maritime urban tracking dataset in harbor environment.

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Marine Navigation

Background:

• Maritime target tracking datasets with ground truth are scarce, hindering the development of algorithms for safe navigation of intelligent marine vessels.
• Existing benchmarks are limited compared to the automotive domain, creating a need for specialized maritime datasets.

Purpose of the Study:

• To introduce the Maritime Urban Tracking (MUT) dataset, designed for perception and tracking tasks in urban water environments for autonomous surface vessels.
• To provide a comprehensive dataset that supports the development and benchmarking of algorithms for maritime autonomous systems.

Main Methods:

• Data collection using an autonomous ferry prototype equipped with stereo cameras, LiDAR, RTK-GNSS, IMU, and a polarized stereo rig.
• Inclusion of target vessels with dual GNSS receivers and Post-Processed Kinematics (PPK) for accurate reference tracks.
• Dataset comprises diverse scenarios including tracking, calibration, mapping, and docking, recorded at high frame rates.

Main Results:

• The MUT dataset includes 19 tracking scenarios, 8 calibration sequences, 1 mapping scenario, and 3 docking scenarios.
• Data covers stereo matching, optical flow, SLAM, 2D/3D object detection, water segmentation, and tracking.
• High-resolution data captured at 30 fps for cameras and 10 Hz for LiDAR.

Conclusions:

• The public release of the MUT dataset aims to reduce entry barriers for researchers and developers in the field of maritime autonomy.
• This resource is expected to foster innovation and accelerate progress in perception and tracking algorithms for autonomous marine vessels.