三光带激光三维人体面部扫描系统的研制

应用物理学论文编号:WLX004  论文字数:26118,页数:71

目    录

第一章 绪论 1
1.1 三维数据获取方法概述 1
1.2面部三维测量的应用 2
1.3面部三维测量的国内外研究状况 3
1.4课题背景 5
1.5论文主要工作及研究意义 5
第二章 扫描系统的原理和总体设计 6
2.1系统的扫描原理 6
2.2系统的总体设计 7
2.2.1 系统的技术指标 7
2.2.2 系统的总体结构 8
2.3系统的硬件设计 9
2.3.1 传感器的设计 9
2.3.2 图像采集卡的选择 11
2.3.3 同步运动系统设计 12
2.3.4 机械结构的设计 13
2.4 系统的软件设计 15
2.5本章小结 18
第三章 系统的标定研究 19
3.1摄像机标定方法概述 19
3.1.1 传统的摄像机标定 19
3.1.2 自标定技术 20
3.2摄像机模型的建立 20
3.2.1 摄像机理想线性模型 20
3.2.2 摄像机实际非线性模型 23
3.3系统的标定及结果 24
3.3.1内参数的标定 24
3.3.2系统的线性标定 26
3.4 本章小结 32
第四章 系统的调试与实验 33
4.1系统的调试 33
4.2 面部模型扫描实验 34
4.2.1 光带中心线提取 34
4.2.2光带分组 34
4.2.3 面部模型点云的获取 36
4.3  点云融合及重建 37
4.3.1直接融合 37
4.3.2 均匀采样融合 39
4.3.3 网格重建 40
4.4 精度验证 41
4.5 本章小结 45
第五章 基于激光扫描的双目立体视觉研究 46
5.1 双目立体视觉传感器 46
5.1.1 摄像机模型 46
5.1.2立体视觉模型 47
5.2 基本矩阵 48
5.3 立体视觉中的极线几何 50
5.4 系统结构的调整 51
5.5系统标定 52
5.6立体匹配 53
5.7 扫描实验 53
5.7.1 实验结果 53
5.7.2 两种方法的比较 55
5.8 本章小结 55
第六章 总结与展望 56
6.1 工作总结 56
6.2对后续工作的建议 56
参考文献 57
发表论文及科研情况说明 61
致    谢 62

摘 要

 本文在全面了解三维人体面部扫描技术国内外研究现状的基础上，基于激光三角法原理，根据分辨率高扫描时间短的要求，提出了三光带激光人体面部扫描方案，设计了系统的软硬件结构，对关键技术进行了研究，并成功研制出系统样机，进行扫描实验得到了预期的实验结果，验证了所设计系统的正确性。主要工作包括以下几个方面：
根据实际使用要求，提出了三光带激光人体面部扫描方案，用三个相同的线激光器产生三条平行且高度方向等间距的光带，扫描时每条光带只需移动整个扫描范围的1/3，满足了缩短系统扫描时间的要求。
设计了扫描系统的总体结构，包括传感器、运动系统、图像采集系统和计算机处理系统，编写了控制程序实现了图像采集系统和运动系统的同步运行，系统的技术指标达到：扫描范围为220mm×250mm×100mm（宽度×高度×深度），典型分辨率为0.4mm×0.4mm×0.5mm（宽度×高度×深度），扫描时间为7s。
建立摄像机理想线性模型和畸变模型得到三维世界坐标和计算机图像坐标的映射关系。将张正友标定算法和摄像机透视变换矩阵标定法相结合对传感器进行标定得到了系统的内参数和外参数。
进行系统调试和实际扫描实验，得到人脸模型的三维点云，通过对点云进行二次采样使摄像机对三条光带的分辨率达到一致，对原始点云和二次采样点云进行重建得到网格模型。对标准量块进行扫描实验得到系统的相对测量误差小于0.25%。
将双目立体视觉原理与主动式激光扫描相结合，提出摄像机固定不动，利用三条线激光进行扫描的方案，获得了人脸点云图和网格模型并对三光带激光扫描和双目视觉两种方法进行了比较。

关键词：人体面部扫描 三光带 传感器标定 二次采样 双目视觉

ABSTRACT
 After analysis of the 3D human face scanning technology at home and abroad, based on laser triangulation principle, in accordance with the requirements of high resolution and short scanning time, a project of three-line structured light laser 3D human face scanning has been proposed. The hardware and software of the system were designed with the key technology studied. The prototype of the system was developed successfully. Scanning experiments were carried out and the expected results verified the correctness of the system. The main work includes:
According to the actual requirements, a project of three-line structured light laser 3D human face scanning was proposed. Three parallel structured light with the same distance in the height direction were projected from three lasers in the same type. When scanning, each laser should move 1/3 of the entire scanning range which meets the requirement of reducing the scanning time of the system.
The general structure of the system were designed which included sensors, motion system, image capture system and computer processing system. The control program has been completed which realized the synchronous running of the image capture system and motion system. The technical indexes of the system reach that: scanning range of 220mm×250mm×100mm(W×H×D), typical resolution of 0.4mm×0.4mm×0.5mm(W×H×D), scanning time of 7 seconds.
By establishing the ideal linear mode and the distortion model of the camera, the mapping relationship between the 3D world coordinates and the computer image coordinates has been deduced. Zhang Zhengyou calibration algorithms and camera perspective transformation matrix calibration method were combined to carry out calibration to the sensor. The intrinsic parameters and external parameters of the system were obtained.
After debugging the system and processing the actual scanning experiment, the point cloud of the face model was obtained. By resampling of the point cloud, the camera resolution of the three-line structured light reached the same. The original point cloud and the resampling point cloud were reconstructed to receive the mesh model. By scanning the standard gauge block, the relative measurement error of the system was less than 0.25%.
By combining the binocular stereo vision and active laser scanning, a project of three-line structured light scanning with the camera fixed was proposed. Point cloud and reconstruction model of the face were obtained. The two methods of three-line structured light laser scanning and binocular vision were compared.

 KEY WORDS：human face scanning, three-line structured light, sensor calibration, resampling, binocular vision