A Multimodal Acousto-Optic Dataset for Underwater Image Enhancement, Detection, and Reconstruction.

Area of Science:

• Marine technology
• Computer vision
• Robotics

Background:

• Underwater exploration relies on acoustic and optical sensing technologies.
• Optimizing multimodal data is crucial for advancing underwater computer vision.
• Existing datasets lack sufficient multimodal data for training and evaluation.

Purpose of the Study:

• To introduce a comprehensive multimodal acousto-optic dataset (MAOUD) for underwater tasks.
• To provide a standardized benchmark for training and validating underwater algorithms.
• To facilitate advancements in underwater image enhancement, detection, and reconstruction.

Main Methods:

• Collected data from a controlled underwater simulation environment.
• Integrated optical RGB images, acoustic images, labeled images, laser point clouds, and acoustic videos.
• Utilized a state-of-the-art underwater multimodal integrated detector for data accuracy validation.

Main Results:

• Developed the MAOUD dataset, a novel resource for underwater research.
• Ensured dataset accuracy through advanced detection and kinematic parameter validation.
• The dataset supports training and evaluation across various underwater acoustic and optical tasks.

Conclusions:

• The MAOUD dataset is vital for developing and benchmarking multimodal underwater algorithms.
• This resource will significantly contribute to progress in underwater exploration technology.
• Standardized datasets are essential for the future of underwater robotics and sensing.