Automatic And Quantitative Data Analysis Methods In Optical Three-dimensional Imaging

Optical-principle-based 3-dimensional (3D) imaging is an important 3D real-world sensing method. Recently, following the rapid development of the computer and opto-electircal technology,3D imaging methods witnessed significant advancements in image resolution, image quality, and image data size. However, after obtaining the high quality 3D image data, how to efficiently and quantitatively analysis it has become a major challenge. This challenge could be interpreted in two different aspects:firstly, in specific imaging applications such as biomedical imaging, the data analysis procedure still highly relies on human perception, which could not meet the growing need of big data analysis. Using computers to achieve automatic data perception has been an irresistible trend. Secondly, the data acquired by 3D optical imaging is always redundant, therefore the most meaningful information should be effectively pull out from the background, during which the quantification of critical parameters of the target 3D object has been involved. Trying to find the solution for these challenges, this thesis studied the automatic and quantitative data analysis methods in optical 3D imaging:First, on the optical 3D image data analysis theory, after reviewing the present general data processing frameworks, we proposed a hierarchical data analysis system that emphasized on the special quantitative measurement needs in the research field of optical engineering. With this hierarchical analysis framework, the chain connecting imaging, image processing and image servoing was constructed.Second, on the calibration of spatial relationships in the 3D imaging data-space, we studied the rectification methods for single and multiple-view geometry and proposed several flexible calibration methods used in stereo vision systems. Moreover, the spatial transformation of 3d image servoing was researched. A quantitative calibration method for the multi-transformation coexist image servoing structure was designed to achieve precise visual feedback between the 3D image data-space and real-world servo mechanism.Third, on the feature extraction of 3D image data, two different types of feature extraction problems were investigated:the first one is on edge detection, a dynamic programming based boundary localization method was designed to achieve stable single line segment extraction; the second one is on region segmentation, we proposed a constrained region growing methodology that incorporates the prior information of specific region properties. Both these new feature analysis approach have obtained comparative significant improvement on different practical 3D image data analysis applications.Fourth, on the automatic and quantitative evaluation of 3D objects, we studied the model structure simplification methods and automatic measurement of critical 3D object structure parameters. Based on this, we proposed a 2D-3D hybrid quantitative analysis method utilizing the spatial projection relationship that unifies the 2D-3D image features. Additionally, we introduced a 3D image data quality assessment method through statistically analyzing 2D feature extraction precision. Using this method, the 3D imaging precision could be quantified using only 2D image processing under fixed imaging hardware.Finally, on the application of automatic and quantitative analysis methods for 3D optical image data, we investigated four different practical real-world scenarios:the flexible laser processing using stereo vision servoing, the quantitative feather material feature evaluation based on trinoculat vision, the automatic airway structure segmentation and thickness measurement in endoscopic optical coherence tomography (OCT) images, and the dynamic quantitative monitoring of hemodynamics in cerebral vasculatures by post-computation of 3D Doppler OCT image data. Multiple original methods and analysis procedures were developed during the research of these applications.To conclude, this thesis has done a relatively comprehensive research on the data analysis problem for 3D optical imaging. The major contribution of this thesis is that new frameworks and noval methods were proposed to tackle the automatic and quantitative 3D image analysis challenge, which greatly improved the efficiency and accuracy of 3D image data processing. To some extent, this thesis has solved certain urgent needs for 3D image perception in the field of optical engineering.