结构光三维扫描测量技术的研究

发布时间：2018-12-18 01:53

【摘要】：本文针对工业现场测量过程中难以快速重构出复杂零件的三维形貌为背景,在结合双目立体视觉技术和结构光测量技术的基础上,深入研究和分析了全视场自定位扫描系统的工作原理及三维重建等问题,提出了一种新的非编码标志圆自定位方法用于重构复杂物体的表面三维轮廓信息,并搭建了硬件系统和测量软件验证了该方法的正确性和可靠性。通过对测量数据进行处理和误差分析,证明了该系统有较高的测量精度和速度。本文主要研究内容和创新成果有:1.提出了一种非编码标志圆自定位技术。创新性的利用标志圆直径唯一性的特征准确识别并区分出极线方向上的多个标志圆,在二维圆准确匹配的基础上,结合空间圆之间的不变拓扑关系,实现同名点云拼接,解决了不同视角下扫描仪空间坐标自标定转换和非连续扫描数据点云测量数据自动拼接难题。2.十字激光线条细化提取及区分。由于线激光光条的高亮特征,可通过自适应阈值分割算法提取出激光条。通过分析现有光条中心提取方法,选取灰度平均加权重心法细化激光线条。普通的十字激光线条区分算法在交叉点提取失败的情况下,将不能做到准确区分,本文在双目系统立体标定和立体校正的基础上,对十字线条进行光平面标定,以唯有正确匹配的光点计算出来的三维点处于光平面上作为区分依据。3.采用Pipeline流水线多线程工作模式,将整个图像处理过程分为四个并行阶段,分别是图像采集、十字激光条细化提取、标志圆识别及匹配、点云重构,调整四个阶段图像处理时间,使得子线程间负载均衡,理论上可将处理时间压缩至原来的四分之一。
[Abstract]:In this paper, it is difficult to quickly reconstruct the 3D topography of complex parts in the process of industrial field measurement. Based on the binocular stereoscopic vision technology and structural light measurement technology, In this paper, the working principle and 3D reconstruction of the whole field of view self-positioning scanning system are deeply studied and analyzed, and a new non-coded marking circle self-localization method is proposed to reconstruct the 3D contour information of complex objects. The hardware system and measurement software are built to verify the correctness and reliability of the method. Through processing and error analysis of measurement data, it is proved that the system has high measuring precision and speed. The main contents and innovations of this paper are as follows: 1. A non-coding marking circle self-location technique is proposed. Innovatively using the unique characteristics of mark circle diameter to accurately identify and distinguish many mark circles in the direction of polar line. On the basis of the accurate matching of two-dimensional circles and the invariant topological relationship between spatial circles, the point clouds of the same name can be stitched together. The problems of self-calibration and self-calibration of scanner spatial coordinate and point cloud measurement of discontinuous scanning data are solved. 2. Cross laser line refinement extraction and differentiation. Because of the highlight feature of the line laser strip, the laser strip can be extracted by adaptive threshold segmentation algorithm. By analyzing the existing methods of extracting the center of the optical strip, the grayscale average plus weight center method is selected to refine the laser line. The common cross laser line discrimination algorithm will not be able to distinguish accurately in the case of cross point extraction failure. Based on the stereo calibration and stereo correction of binocular system, the cross line is calibrated in optical plane. The three dimensional points calculated by only the correct matching points are located on the optical plane. 3. 3. The whole image processing process is divided into four parallel stages by using Pipeline pipeline multithreading mode, which are image acquisition, cross laser strip thinning extraction, mark circle recognition and matching, point cloud reconstruction, and adjustment of image processing time. This allows for load balancing between child threads, which theoretically compresses processing time to the original 1/4.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】