结合法向信息的立体匹配研究

发布时间：2018-06-13 07:53

本文选题：立体匹配 + 深度置信度　；参考：《浙江大学》2017年硕士论文

【摘要】：深度恢复是计算机视觉领域的基本问题,有着广泛的应用。.其中,基于双目相机的立体匹配是最常用的深度恢复方式之一,有着重要的理论研究和实际应用价值。传统的双目立体匹配算法在处理弱纹理和遮挡区域时,容易因为匹配困难而导致难以准确地获得其视差,也因此较难恢复出高质量的深度。为了克服这些因为自身匹配信息不足而导致难以准确获得视差的情况,本文提出了一种结合法向信息预测的立体匹配视差恢复方法,通过求解同时含有法向和视差约束项的最小二乘线性方程组,将单张图像的法向估计信息转化为视差信息。为了达到这个目的,本文提出了基于平面假设的视差置信度估计算法来衡量视差值的可靠性,从而为视差求解提供稀疏的可靠视差点,作为对视差绝对数值的约束。此外,还提出了针对视差跳变区域的边界检测和融合算法,以准确地获得视差不连续处的位置信息,从而能在视差平滑区域内维持法向对相邻像素视差大小和变化的约束;在视差跳变处不使用法向约束,以保持视差的不连续性。最后对求得的视差图进行迭代优化,在每一轮迭代中通过中值滤波补全视差图,并优化每一轮中的边缘信息和视差约束点集。针对实时应用需求,本文也提出了快速的版本,可以达到每秒一帧以上的计算速度。实验结果表明,本文提出的方法在对公开的测试数据集和视频序列进行立体匹配的过程中,均能够有效地提高弱纹理和遮挡区域的视差质量和深度恢复精确度。该方法具有较好的扩展性和普适性,可以和多种立体匹配算法结合来提高深度恢复的质量。
[Abstract]:Depth recovery is the basic problem in the field of computer vision. It has extensive application.. among them, stereo matching based on binocular camera is one of the most commonly used depth recovery methods. It has important theoretical research and practical application value. The traditional binocular stereo matching algorithm is difficult to match the weak texture and occlusion area because it is difficult to match. It is difficult to obtain the parallax accurately, and therefore it is difficult to recover the high quality. In order to overcome the difficulty of obtaining the parallax precisely because of the insufficient matching information, this paper proposes a stereo matching parallax method, which combines the normal and the parallax constraints. In order to achieve this goal, this paper proposes a parallax reliability estimation algorithm based on plane hypothesis to measure the reliability of parallax values, thus providing a sparse reliable parallax point for parallax solution as a constraint on absolute values of parallax. In addition, a boundary detection and fusion algorithm for parallax hopping region is proposed to obtain accurate location information of parallax discontinuities, so as to maintain the constraint on the parallax size and change of the adjacent pixels in the parallax smooth region, and keep the discontinuity of parallax in the parallax jump to maintain the discontinuity of the parallax. The parallax graph is iteratively optimized, and the parallax graph is complementing through median filtering in each round, and the edge information and the parallax constraint point set in each round are optimized. In view of the real-time application requirements, a fast version is also proposed in this paper, which can reach the calculation speed of more than one frame per second. The experimental results show that the method proposed in this paper is used in this paper. In the process of stereo matching of open test data sets and video sequences, the quality of parallax and depth recovery in the weak texture and occlusion region can be effectively improved. The method has good scalability and universality, and can be combined with a variety of stereo matching algorithms to improve the quality of the depth recovery.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】