【作者】 李响；

【导师】 王鹏飞；

【作者基本信息】 哈尔滨工业大学 ， 机械工程（专业学位）， 2019， 硕士

【摘要】 海洋作为地球物种起源的摇篮,占据着地表面70.8%的空间,是维护地球生物多样性的重要保障。随着世界人口数目的不断增长以及内陆可利用资源的日益匮乏,人类意识到海洋内蕴藏的丰富生物种类及矿藏种类将成为未来人类生存资源的保障。目前海洋受到的污染非常严重,海上的垃圾、漂浮物等不断增加,并且数量已经远超海洋的净化能力。由于人体的生理特征不适合于长时间的海洋环境作业,因此,浅海水下的打捞设备应运而生。但目前打捞设备效率低下,自动化程度不高,因此,本文通过计算机视觉对基于浅海水下环境的目标物检测与位姿估计进行了研究。首先,针对水下悬浮物对光散射导致的水中雾化问题,基于陆上暗通道先验的去雾算法,提出了一种基于水下环境的背景光估计方法,融合了暗通道先验估计系数和在水下衰减较弱的蓝、绿通道系数,实现了对水下图像的复原;针对光在水中严重衰减导致的图像对比度下降、颜色失真和细节模糊等问题,对直方图均衡化和拉普拉斯锐化两种算法进行了融合,提出了一种自适应增强算法,使对不同环境的图像得到合理的增强。同时对处理后的图像进行了客观质量评价。其次,针对水的折射造成的位置偏移问题,基于Snell定律建立了水下双目相机成像模型,提出了一种以固有深度为先验的深度矫正方法,完成了对水下位姿的矫正及实验验证;面对基于特征点的普适性匹配算法在水下环境的识别率低、匹配速度慢等一系列问题,本文应用了一种基于识别码的目标物检测,并提出了一种应用于空间曲面的水下识别码识别算法,通过优化大大提高了识别速度,通过实验验证了算法可行性;针对水下成像质量差,点估计和线估计无法满足精度要求的问题,将RANSAC点云分割和ICP点云配准的方法应用到水下位姿估计。通过光学式运动捕捉系统建立了评价标准,对双目相机位姿估计精度进行了评价。最后,搭建实验平台,模拟水下环境条件,通过实验验证了水下图像处理和目标物检测与位姿估计方法的有效性。基于识别码对水下各算法处理后的图像进行目标检测,验证了水下图像处理算法的必要性和优越性;通过室外目标物检测与位姿估计实验验证了目标物检测与位姿估计算法的可行性及识别率。更多还原

【Abstract】 As the cradle of earth’s species origin,the ocean occupies 70.8 percent of the earth’s surface space,it is an important guarantee for the sustainable biodiversity of our planet.With the rapid growth of the world’s population and the increasing scarcity of available resources in the inland,humans realize that the rich biological and mineral species in the ocean will become the guarantee of future survival resources.At present,the ocean is seriously polluted by human beings,and the amount of garbage and floating objects on the sea is increasing,which has far exceeded the purification capacity of the ocean.And because the physiological characteristics of the human body are not suitable for long-term Marine environment operations,therefore,the shallow underwater fishing equipment came into being.At present,the efficiency of salvage equipment is low and the degree of automation is not high.Therefore,this paper studies the detection and pose estimation of target based on the shallow underwater environment through computer visionFirstly,in view of the problem of water atomization caused by the scattering of light from underwater suspended material,based on the defog algorithm of land-based dark channel prior to(DCP),a background light estimation method based on the underwater environment is proposed,which combines the coefficient estimation of the dark channel and the blue and green channel with weak underwater attenuation,and realizes the restoration of the underwater image.In view of the problems of image contrast drop,color distortion and detail blur caused by the severe attenuation of light in water,the histogram equalization and Laplace sharpening are combined,and an adaptive enhancement algorithm is proposed to make the image of different environments be reasonably enhanced.At the same time,the processed image is evaluated objectively.Secondly,in view of the location offset caused by the refraction of water,based on Snell’s law,an underwater binocular camera imaging model is established,a depth correction method with inherent depth as a priori is proposed,and the correction and experimental verification of the underwater position are completed.Facing with a series of problems such as low recognition rate and slow matching speed of feature point-based universal matching algorithm in underwater environment,and proposed an underwater identification code recognition algorithm applied to the surface,through optimization,the recognition speed is greatly improved,and the feasibility of the algorithm is verified by experiments.The RANSAC point cloud segmentation and ICP point cloud registration were applied to the underwater pose estimation because of the poor underwater image quality and the inability of point estimation and line estimation to meet the accuracy requirements.Finally,an evaluation standard is established by the optical motion capture system to evaluate the accuracy of the pose estimation of binocular camera.Finally,an experimental platform was built to simulate the underwater environment conditions,and the effectiveness of underwater image processing,the target detection and pose estimation is verified by experiments.The necessity and superiority of underwater image processing algorithm are verified by target detection of images processed by different algorithms based on recognition code.The feasibility and identification rate of the target detection and pose estimation algorithm are verified by outdoor the target detection and pose estimation experiments.更多还原