毕业生知识库

近年来，三维重建与检测技术有着广泛的应用，金属材质产品的三维重建研究也存在极大需求，然而高光表面的镜面反射、互反射特性并不满足大部分光学重建方法的前提要求，其重建问题依旧是一个难点。在已有研究中，一般采用基于时间编码法的结构光法获取高光表面的高分辨率重建结果，以及采用曝光融合的方法避免曝光过度、曝光不足等不理想区域出现，但此类方法都会导致重建时间成倍增长。可见完成高分辨率、高精度、高速度的高光表面三维重建依旧存在许多障碍，如何解决这些问题需要进一步的研究。
针对以上问题，本文围绕实际工业检测场景需求，以实现分辨率、精度、速度、成本、易用性均有优秀表现的高光表面重建为目标，设计具有创新性的彩色格雷码结构光模式图，并优化基于像素点质量的曝光融合方法，开展对高光表面的三维重建研究。主要研究工作和贡献归纳如下：
（1）提出了结合彩色信息与时间编码法的彩色格雷码结构光法，能够有效提升三维重建速度和点云稠密度。本文针对时间编码法耗时长的问题，加入彩色编码信息以减少模式图张数，并通过双目相机的使用解决相机-投影仪色彩矫正的难点，以格雷码思想降低误匹配代价，投射逆向模式图以辅助像素点色彩值的判断，并尝试构建彩色线移法模式图提高分辨率，最终在成功缩减近三分之二投影时长的同时提升了重建精度。此方法充分考虑了非专业投影仪的特点、避免了投影仪的标定过程，因此同样是低成本、易实现的。
（2）提出了基于直接式曝光融合方法和彩色结构光的高光表面三维重建方法，能够快速将曝光不理想的图像序列融合为高曝光质量的图像，作为输入进行正常的三维重建。针对曝光融合速度慢、效果差的问题，本文避免计算高动态范围的中间结果，而是以基于像素点质量完成融合的方法进行加速，通过多次实验缩短曝光时间序列；同时结合前文提出的彩色结构光特点优化融合参数，进一步加速曝光融合过程、提升融合质量，在三维重建结果中有较好的表现。
（3）以工业场景下的小尺寸高光表面的三维检测问题为背景，设计实现了三维重建系统。本文以实际工业场景中锂电池缺陷检测为初始背景与需求，对所需的相机、投影仪产品的分辨率、帧率、焦距等参数进行分析计算，平衡精度、速度、成本上的需求，进行设备的选型。结合必要的辅助工具，同时集成软件模块，搭建出针对尺寸较小、漫反射或镜面反射被测表面的三维重建系统，提出完整的三维重建方法与流程。实验中初步验证了重建系统与流程的有效性，完成原型演示。

In recent years, three-dimensional reconstruction and detection technology have been widely used. The three-dimensional reconstruction of metal materials products is also in great demand. However, the specular reflection and mutual reflection characteristics of shiny surfaces do not meet the prerequisite requirements of most optical reconstruction methods. Three-dimensional reconstruction of shiny surfaces with high quality is still a difficult problem. In the existing research, structured light methods with time coding are widely used to obtain high-resolution results. In the meantime, exposure fusion methods are often used to avoid undesirable exposure, avoiding over-exposure and under-exposure areas. But these methods will lead to several times longer reconstruction time. Therefore, there are still many obstacles to complete three-dimensional reconstruction with both high-precision and high-speed on shiny surfaces.
To solve the above problems, this article focuses on the needs of actual industrial detection scenarios, with the goal of achieving shiny surface reconstruction with excellent performance in resolution, accuracy, speed, cost, and ease of use. We designed innovative colored gray code structured light patterns, and optimize the fusion method of exposure based on the quality of pixels. In the end, we carried out researches on three-dimensional reconstruction of shiny surfaces. Main research contents and contributions are summarized as follows:
(1) A method of colored gray code structured light combining color information and time coding methods was proposed, which can effectively improve the speed of three-dimensional reconstruction and the density of point clouds. Considering how time-consuming  the time coding methods are, we added color information to reduce the number of the structured light patterns. The difficulty of camera-projector color correction was solved using the binocular camera. The idea of Gray code was also used to reduce the cost of mismatching. We projected inverse pattern to assist the judgment of the color value of each pixel. We also attempted to propose a colored line-shifting method to gain better resolution. Finally, this method successfully reduced the projecting time and improved the reconstruction accuracy. At the same time, this method fully considered the characteristics of non-professional projectors, and avoided the calibration procedure of the projectors, which means it is also low-cost and easy to implement.
(2) A three-dimensional reconstruction method of shiny surface based on direct exposure fusion methods and colored structured light was proposed, which can quickly fuse the image sequence with poor exposure quality into a high-exposure quality image as input for normal three-dimensional reconstruction. With exposure fusion methods often being slow and have poor effects, in this paper calculating the intermediate results of high dynamic range was avoided. On the contrast, the fusion method was accelerated based on pixel quality. Meanwhile, the exposure time sequence was shortened through multiple experiments. This method combined the colored structured light method proposed above. At the same time, the fusion parameters were optimized based on the characteristics of the colored structured light method proposed above, which further accelerated the exposure fusion process and improved the fusion quality, and had a better performance in the 3D reconstruction.
(3) Based on the problem of three-dimensional detection of small-sized shiny surfaces in read industrial scenarios, a three-dimensional reconstruction system was designed and implemented as well. This paper took the detection problem of lithium battery defections in actual industrial scenarios as the initial background and requirements. We analyzed and calculated the required resolution, frame rate, focal length and other parameters of the camera and projector products to meet and balance the needs of accuracy, speed and cost Equipment selection and purchase were carried. Combining the necessary auxiliary tools and integrating software modules at the same time, a 3D reconstruction system for the measured surface with small size, diffuse reflection or specular reflection was built, and a complete 3D reconstruction method and process was proposed. The experiment had completed the prototype demonstration and initially verified the effectiveness of the reconstruction system and process.