基于双目立体视觉的水下高精度三维重建方法

发布时间：2018-04-13 16:07

本文选题：相机标定 + 成像模型　；参考：《哈尔滨工业大学》2017年硕士论文

【摘要】：随着人类经济社会发展,石油、天然气等资源的消耗与日俱增,陆地上这些资源正被开发殆尽。而占地球表面积三分之二的海洋中仍然有大量资源尚未被开发。由于深海环境对人类相当危险,因此对海洋的探索主要依靠水下机器人。利用视觉信息对周围环境进行三维重建,对水下机器人的导航及探索活动具有重要应用价值,也是目前计算机视觉领域的重要研究方向。但由于水下环境较为复杂,水下照片质量下降,并且会发生畸变,使得传统空气中相机标定及三维重建算法不能直接应用于水下。本文针对课题内容及算法要求,对空气中三维重建算法进行了深入研究并改进,以适应水下重建的需要。采用多分辨率立体匹配结合多种约束条件,得到视差图。并且针对目前常用方法中存在的模型阶梯阴影误差问题,提出了一种新的视差图优化方法,利用各向异性滤波处理视差图,使算法的精度及鲁棒性得到了提升,同时运行时间也有所减少,提高了算法效率。针对水下成像模型与空气中模型的区别,对水下相机进行标定并且对存在的像素偏移误差进行补偿。为了研究水下成像过程,本文对理想情况下水下成像过程进行建模,同时改进一种基于平面标定板的水下相机标定方法,将折射参数加入到标定模型中,完成水下相机标定。根据实际实验条件与理想条件不同,建立非理想条件下水下图象复原模型,引入像素偏移误差参数,设计算法对其进行补偿,并进行实验验证补偿效果。补偿后的水下图像即可复原为空气中图像,并且应用空气中的双目重建算法,得到水下物体的三维模型。本文设计了多组实验,分别对多个水下物体进行建模,并且把水下重建结果与空气中的结果进行对比。相对于空气中的重建模型,本课题所提出的算法,有重建精度高、运行效率高、模型完整度好的优点,可以应用在实际水下研究工作中。
[Abstract]:With the development of human economy and society, the consumption of oil, natural gas and other resources is increasing day by day.The oceans, which account for 2/3 of the earth's surface area, still have vast untapped resources.Because the deep sea environment is very dangerous to human beings, the exploration of the ocean mainly depends on underwater vehicles.Using visual information to reconstruct the surrounding environment has important application value for underwater vehicle navigation and exploration activities. It is also an important research direction in the field of computer vision.However, due to the complexity of underwater environment, the quality of underwater photos is decreased and distortion occurs, the traditional camera calibration and 3D reconstruction algorithms in the air can not be directly applied to underwater.In order to meet the need of underwater reconstruction, the 3D reconstruction algorithm in air is studied and improved in this paper.The parallax map is obtained by multi-resolution stereo matching combined with many constraints.In order to solve the problem of model step shadow error, a new parallax map optimization method is proposed, which can improve the accuracy and robustness of the algorithm by using anisotropic filter to process the parallax map.At the same time, the running time is also reduced, and the efficiency of the algorithm is improved.Aiming at the difference between underwater imaging model and airborne model, the underwater camera is calibrated and the pixel offset error is compensated.In order to study the underwater imaging process, the underwater imaging process is modeled in this paper, and an underwater camera calibration method based on the plane calibration plate is improved. The refraction parameters are added to the calibration model to complete the underwater camera calibration.According to the difference between the actual experimental conditions and the ideal conditions, the underwater image restoration model under the non-ideal condition is established. The pixel offset error parameter is introduced, the algorithm is designed to compensate it, and the compensation effect is verified by experiments.The compensated underwater image can be restored to the image in the air, and the 3D model of underwater object can be obtained by using the binocular reconstruction algorithm in the air.In this paper, several experiments are designed to model several underwater objects, and the results of underwater reconstruction are compared with those in air.Compared with the reconstruction model in air, the algorithm proposed in this paper has the advantages of high reconstruction accuracy, high running efficiency and good model integrity, so it can be used in practical underwater research.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】