Multimodal Underwater Object Detection and Segmentation Using MOUD (RGB + LiDAR) Point-Cloud Dataset

View at publisher Summarise with VIRAT
Authors: Revathy D, R. Swathika, N Radha
Cited by: 0
References: 14

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Underwater image enhancement is critical for improving the performance of marine applications such as archeological exploration, autonomous navigation, surveillance, and underwater object detection. However, underwater images often suffer from severe degradation caused by wavelength-dependent light absorption, scattering, and artificial illumination, resulting in reduced contrast, color distortions, hazy appearance, and loss of structural details. While traditional enhancement methods such as histogram equalization and physical model–based correction—address basic distortions, modern deep learning techniques, including CNN- and GAN-based models, provide superior reconstruction capabilities through deep feature extraction. This paper aims to develop a hybrid enhancement framework that integrates classical underwater correction methods with advanced deep learning techniques (ADLT) to improve raw underwater RGB image quality. A specialized underwater-aware loss function will be designed to handle color imbalance, haze removal, and structural consistency. To ensure compatibility with modern multimodal systems, the enhanced RGB images will be aligned with the MOUD dataset, a multimodal RGB + LiDAR point-cloud benchmark used for underwater object detection and segmentation. Enhancing RGB images before multimodal fusion is expected to significantly improve the accuracy of object recognition, depth estimation, and segmentation tasks in challenging underwater environments. By combining conventional preprocessing, deep learning–based enhancement, and multimodal alignment with MOUD, the proposed approach is expected to deliver visually superior and analytically reliable underwater images that enhance the performance of real-time underwater robotics and sensing applications. The proposed ADLT method achieves 73% mAP, 67% mIoU, 31 PSNR, and 36 ms inference time, demonstrating superior detection, segmentation, enhancement quality, and real-time efficiency.

Fields of Study & Subjects

Image Enhancement Techniques Underwater Vehicles and Communication Systems Advanced Neural Network Applications

Publication details

Journal / source
International Conference on Computing, Communication, Control and Cyber-Physical Systems (I5CPS)
Pages
1–8
Publication date
26 February 2026
Language
EN