立体匹配技术在波浪摄影测量中的应用研究

发布时间：2019-02-24 18:05

【摘要】：本文以双目摄影波浪测量技术为背景,以增加特征点匹配对数目、降低误匹配率为目的,研究双目视觉里立体匹配技术的特征提取与特征匹配,并将改进的立体匹配算法应用于水槽波浪的浪高测量。首先,介绍经典的特征提取算法Moravec、Harris与SIFT,并进行仿真实验与波浪图像的特征提取,实验结果表明Harris角点分布更能反映波浪结构,在波浪图像发生多种变化时SIFT特征提取的稳定性最好。接着,介绍金字塔匹配与SIFT匹配,并分别进行仿真实验与波浪图像的特征匹配,实验表明SIFT匹配处理波浪图像时误匹配率不高,但匹配对数目少。然后,针对SIFT算法处理波浪图像时匹配对数量少、误匹配对等缺点,本文提出了一种改进的立体匹配算法。特征提取部分,采用Harris亚像素角点替代DoG极值点,通过改变Harris角点响应阈值改变特征点数量。Harris亚像素角点是通过二次多项式逼近角点响应值的方法得到的。特征匹配部分,采用双向匹配策略剔除误匹配对,在双向匹配策略中采用极线约束确定候选匹配集。实验结果表明,本文的改进算法与传统SIFT算法相比,不仅误匹配率较低,而且匹配对数量有了明显增加。最后,为了验证本文立体匹配改进算法在波浪图像上的可应用性,采用本文的改进算法处理室内实验与循环水槽实验采集的波浪图像,并将匹配结果用于三维重建得到波浪的三维立体图与等高线图。实验结果表明,三维立体图的波形、波向与实际波浪的相同;水槽实验中标记处的两组波高测量值与实际值误差分别约为2.67%与4.84%。
[Abstract]:In this paper, the feature extraction and feature matching of stereo matching in binocular vision are studied in order to increase the number of matching pairs of feature points and reduce the mismatch rate in the background of binocular photowave measurement. The improved stereo matching algorithm is applied to wave height measurement of flume. Firstly, the classical feature extraction algorithms Moravec,Harris and SIFT, are introduced, and the simulation experiments and wave image feature extraction are carried out. The experimental results show that the Harris corner distribution can better reflect the wave structure. The stability of SIFT feature extraction is the best when the wave image changes. Then, the pyramid matching and SIFT matching are introduced, and the simulation experiments and the wave image feature matching are carried out respectively. The experiment shows that the mismatch rate is not high, but the number of matching pairs is small. Then, an improved stereo matching algorithm is proposed to solve the problem that the SIFT algorithm has fewer matching pairs and less mismatch when processing wave images. In the feature extraction part, the Harris sub-pixel corner is used to replace the DoG extremum, and the Harris corner response threshold is changed to change the number of feature points. The Harris sub-pixel corner is obtained by approximating the corner response value by quadratic polynomial. In the part of feature matching, bidirectional matching strategy is used to eliminate mismatch pairs, and polar line constraints are used to determine candidate matching sets in bidirectional matching strategies. The experimental results show that compared with the traditional SIFT algorithm, the improved algorithm not only has a lower mismatch rate, but also has a significant increase in the number of matching pairs. Finally, in order to verify the applicability of the improved stereo matching algorithm in wave images, the improved algorithm is used to deal with the wave images collected from indoor experiments and circular flume experiments. The matching results are used to reconstruct the wave images and contour images. The experimental results show that the waveforms of the three-dimensional images are the same as those of the actual waves, and the errors between the measured values and the actual values of the two groups of wave heights in the flume experiment are about 2.67% and 4.84%, respectively.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U661.7;TP391.41

【参考文献】