真实海洋环境中水下机器人视觉图像清晰化方法研究

In real complex marine environment applications, good environmental perception is the key to ensure the operation ability and intelligence of underwater robots. Focusing on the series of problems existed in conventional underwater image enhancement, such as limitations in optical imaging model, non-uniform illumination introduced by artificial light, ill-posed estimation of multi-parameter, and noise amplification, this project tries to carry out research on robust underwater image enhancement method for the real operating environment of underwater robots. The main contents include: An optical imaging model for underwater robots in real marine environment is conducted. A non-uniform illumination compensation and removal method is explored. A single parameter based multi-scale gradient domain underwater image enhancement method is proposed. Additionally, the above research results are integrated into the experimental platform of the underwater fishing robot, which is developed by our research group, to verify the validity of the methods. The research results of this project are expected to improve the environmental perception of underwater robots in real marine environment and ensure the accuracy of object recognition. It has significant application value to realize the autonomous and accurate fishing of underwater robots, and is of great economic and social significance to promote the upgrading of the marine industry and the prosperity of the marine economy.

在真实复杂海洋环境应用中，良好的环境感知能力是确保水下机器人作业能力和智能化水平的关键。针对目前水下图像清晰化中存在的光学成像模型局限性、主动照明引入非均匀光照、多参数病态估计、噪声放大等一系列问题，本项目将开展面向水下机器人真实作业环境的具有鲁棒性的水下图像清晰化方法研究。主要内容包括：构建真实海洋环境中面向水下机器人的光学成像模型，探索非均匀光照的补偿去除方法，提出单一参数的多尺度梯度域水下图像增强方法，并将上述研究成果集成至课题组研发的水下捕捞机器人样机实验平台中进行验证。本项目的研究成果有望提高真实海洋环境中水下机器人环境感知能力，确保后续目标识别的准确性，对实现水下机器人自主、精确捕捞具有显著应用价值，对促进海洋产业装备升级和海洋经济繁荣具有重要的社会经济意义。

针对目前水下图像清晰化中存在的光学成像模型局限性、主动照明引入非均匀光照、多参数病态估计、噪声放大等一系列问题，本项目以水下机器人真实作业环境为研究背景，旨在建立具有鲁棒性的水下图像增强系统，提高水下机器人在复杂海洋环境中的感知能力，解决水质浑浊及主动照明带来的诸多问题。重点研究内容包括：构建真实海洋环境中面向水下机器人的光学成像模型，模拟水下机器人光学成像过程，克服传统模型局限性；探索非均匀光照的校正补偿方法，去除主动照明引入的非均匀光照，以便对分离出的均匀光照背景区域进行有效增强；提出单一参数的多尺度梯度域水下图像增强方法，避免原物理模型多参数估计误差导致的失真，并可在增强细节的同时有效抑制噪声；最后将上述研究成果集成至课题组研发的水下捕捞机器人样机实验平台中进行验证。.课题组按照计划，围绕以上目标开展研究工作，在水下光学成像模型构建、非均匀光照校正以及多尺度梯度域水下图像清晰化等具体应用方面均取得突破。相关工作共发表已标注的论文12篇，公开发明专利9项，部分研究成果已部署在课题组自主研发的水下捕捞机器人平台上进行实际应用，且已获得优异效果。研究成果有望提高真实海洋环境中水下机器人环境感知能力，对实现水下机器人自主、精确捕捞具有显著应用价值，对促进海洋产业装备升级和海洋经济繁荣具有重要的社会经济意义。