Remote and Accurate Detection of Underwater Obstacles with Novel Laser Structure Light Scotopic Vision and AI Machine Learning

Abstract

Considering the high-cost but poor structure, shape and texture features of the sonar systems, a cost-effective laser-based structur e light (LbSL) scotopic vision system is proposed to capture the high-quality images in distance for more accurate detection and cla ssification of underwater obstacles. By combining LbSL with Diffractive Optical Element (DOE), scotopic vision and AI machine learni ng, various cutting-edge techniques from image acquisition, enhancement, feature extraction and analytics to deep learning based int erpretation, we aim to solve a series of technical challenges. These include i) Poor resolution and high cost for imaging the object s of interest (e.g. sonar, LiDAR and LLS); ii) Short working range (i.e. 0-3 m) for most optic systems; iii) No available solutions for reliable and accurate detection and classification of underwater obstacles. As a result, the detection range, environment and ac curacy can be significantly improved, which can further benefit refined decision-making to tackle the detected obstacles, including optimised recommendations e.g. path-planning for smoother and more autonomous operations of naval missions and other maritime tasks. The key objectives are summarized as follows:1) To develop a LbSL based laser-optic trigonometric imaging system for remote sensing of challenging underwater scenes, with improved working range and resolution in point-cloud and visible images;2) To fuse both visib le image and point-cloud data in a hybrid framework with DL-enabled detection and classification of underwater obstacles;3) AI ML ba sed smart decision-making for tackling of detected obstacles.Major outcomes include i) a demonstrable prototype, ii) relevant algori thms/software tools, iii) patents, reports/publications where applicable, and iv) established team with good links to naval communit y. Some dissemination actions are summarized as follows:1) A project webpage for regular updates;2) A one-day networking event in o ur National Subsea Centre to showcase any results/findings;3) Onsite demo as arranged by the funder;4) Release of project software/t ools/datasets & reports/publications.

Document Details

Document Type
DoD Grant Award
Publication Date
Sep 07, 2021
Source ID
N629092112057

Entities

People

  • Jinchang Ren

Organizations

  • Office of Naval Research
  • The Robert Gordon University
  • United States Navy

Tags

Readers

  • Computer Vision.
  • Marine Propulsion Engineering and Naval Architecture
  • Snow Cover Descriptors for Reptiles and Their Illustrations.

Technology Areas

  • AI & ML
  • AI & ML - Autonomous Systems
  • AI & ML - DoD AI Strategy
  • Directed Energy