基于激光散斑的深度图获取算法研究.pdf-资料库

df53b09d-f779-456d-8d94-7aad198ddb73.pdf-第1页.png

第1页 / 共66页

df53b09d-f779-456d-8d94-7aad198ddb73.pdf-第2页.png

第2页 / 共66页

df53b09d-f779-456d-8d94-7aad198ddb73.pdf-第3页.png

第3页 / 共66页

df53b09d-f779-456d-8d94-7aad198ddb73.pdf-第4页.png

第4页 / 共66页

df53b09d-f779-456d-8d94-7aad198ddb73.pdf-第5页.png

第5页 / 共66页

df53b09d-f779-456d-8d94-7aad198ddb73.pdf-第6页.png

第6页 / 共66页

df53b09d-f779-456d-8d94-7aad198ddb73.pdf-第7页.png

第7页 / 共66页

df53b09d-f779-456d-8d94-7aad198ddb73.pdf-第8页.png

第8页 / 共66页

分类号:TP301.6 10710-2014124046 硕士学位论文基于激光散斑的深度图获取算法研究侯旭阳导师姓名职称申请学位类别张绍阳教授工学硕士学科专业名称计算机应用技术论文提交日期 2017 年 5 月 27 日论文答辩日期 2017 年 6 月 4 日学位授予单位长安大学

Study on the Algorithm of the Depth Image’s Acquiring Based on Laser Speckle A Thesis Submitted for the Degree of Master Candidate：Hou Xuyang Supervisor：Prof. Zhang Shaoyang Chang’an University, Xi’an, China II

论文独创性声明 本人声明：本人所呈交的学位论文是在导师的指导下，独立进行研究工 作所取得的成果。除论文中已经注明引用的内容外，对论文的研宄做出重 要贡献的个人和集体，均己在文中以明确方式标明。本论文中不包含任何 未加明确注明的其他个人或集体己经公开发表的成果。 本声明的法律责任由本人承担。 论文作者签名：年４月沴曰 ＞〇门论文知识产权权属声明 本人在导师指导下所完成的论文及相关的职务作品，知识产权归属学 校。学校享有以任何方式发表、复制、公开阅览、借阅以及申请专利等权 利。本人离校后发表或使用学位论文或与该论文直接相关的学术论文或成 果时，署名单位仍然为长安大学。 （涉密论文在解密后应遵守此规定） 论文作者签名：ｆｓ年／月爷曰 导师签名：＾年７月３■ 日 ．〈ｒ

摘要深度图像是指将从图像采集器到场景中各点的距离作为像素值的图像，它能有效的区分前景和背景，对于目标识别更加高效。本文研究了现有的深度成像的成像原理和国内外能达到量产的深度相机，目前国外已经有实时的检测大范围、含有运动场景、且成本低的深度成像系统。由于技术封锁，国内的类似设备性能相对还较差。为此，本文对基于激光散斑的深度图获取算法进行了深入的研究。首先，本文对激光散斑图像的成像原理、成像系统、散斑成像光栅进行了详细的研究，对散斑图像的自相关性和互相关性进行了实验分析，对其作为空间结构光来进行深度检测的可能性进行了研究。接着，本文对空间结构光获取深度的原理进行了研究，通过实验分析了深度与散斑图像的关系，推导了深度图的计算公式。提出了一种高效的、简便的块匹配方法，快速的获取深度图，详细叙述了使用这种快速块匹配方法获取整幅图像深度的具体流程。然后，本文对深度图像获取算法进行了优化，包括速度和精度的优化。在速度方面，研究了匹配块的大小的选择；提出了一种搜索块大小动态可变的搜索方法；同时提出了一种快速的匹配策略，大大降低计算量。在精度方面，在研究了散斑成像系统参数对精度的影响因素的基础上，给出了从散斑成像系统参数上改善精度的方法；提出了一种横向偏移量插值的插值算法以提高深度摄像头的深度分辨率。最后，通过实验表明：本文所使用的块匹配方法，在本文第三章所述散斑成像系统参数下，深度精度在 50cm 处可以达到 1cm，在 200cm 处可以达到 5cm，可以满足大部分的检测要求。在速度方面，把散斑图像传到处理器为双核 3.6Ghz，内存为 4GB 的 PC 机上，在 PC 机上进行的实验，计算一幅 1280*720 的深度图，时间可以达到 0.5s 以内。关键词：激光散斑，结构光，深度图，快速匹配方法 i

Abstract Depth image means the value of pixels in the image is the distance between the object and camera. It can distinguish the foreground and background effectively, so it is more efficient for the target recognition. This paper studied the principle of depth camera and the cameras that can reach mass production. At present there are already the depth camera that can detect a wide range, contains motion scene in real time, and the cost is low abroad. Due to technical blockade, similar equipment’s performance is relatively poor domestic. In order to improve the performance, this paper studied the method of acquiring of depth image based on laser speckle. Firstly, this paper studied the principle, system and optical grating of speckle, analyzed the self-correlation and cross-correlation of speckle through experiment. Then this paper studied the possibility of the speckle image as a space structured light to detect the depth. Secondly, this paper studied the principle of the depth image’s acquiring through space structured light, analyzed the relationship between depth image and speckle image，derived the formula of depth image. This paper proposed an efficient and easy block matching method, which can acquire depth image quickly, narrated the process of acquiring the whole image’s depth using the way of exclusive or block matching detailedly. Thirdly, this paper optimized the arithmetic of the depth image acquiring, both on speed and accuracy. In terms of speed, this paper studied the best choice of size of matching block, proposed a variable searching block size searching method, proposed a fast matching strategy, to reduce the calculation greatly. In terms of accuracy, this paper studied the influence of speckle image system’s parameter and introduce the way to improve the accuracy through adjust the speckle image system’s parameter. Then this paper proposed a method of improving the accuracy, an interpolation method of crosswise offset interpolation to improve the depth camera’s resolution. Finally, the experimental results show that the block matching this paper proposed, with the speckle image system’s parameter showed in chapter three of this paper, depth accuracy can reach 1cm in the distance of 50cm and 5cm in the 200cm. This can satisfy the detection requirements for most scenes. In terms of speed, put the speckle images to personal computer ii

with 3.60Ghz dual-core processor CPU and 4GB RAM configuration, and then do experiment on the PC, we can get a 1280 * 720 pixels depth image within 0.5 seconds. Keywords: Laser speckle, Structured light, Depth image, Fast matching method iii

目录第一章绪论 ........................................................................................................ 1 1.1 研究背景和意义 .......................................................................................................... 1 1.2 深度成像方法研究 ...................................................................................................... 2 1.2.1 被动式成像方法 .............................................................................................. 3 1.2.2 主动式成像方法 .............................................................................................. 5 1.2.3 不同的深度检测方法的比较 .......................................................................... 8 1.3 国内外研究现状 .......................................................................................................... 9 1.3.1 国外研究现状 .................................................................................................. 9 1.3.2 国内研究现状 ................................................................................................. 11 1.4 本文的主要研究内容 ................................................................................................ 12 1.5 本文结构安排 ............................................................................................................ 13 第二章散斑的产生与特性分析 ...................................................................... 14 2.1 激光散斑理论基础 .................................................................................................... 14 2.1.1 激光散斑产生原理 ........................................................................................ 14 2.1.2 激光散斑测量方法 ........................................................................................ 15 2.2 散斑空间结构光的产生 ............................................................................................ 16 2.3 散斑空间结构光的特性分析 .................................................................................... 19 2.3.1 散斑图像自相关特性分析 ............................................................................ 20 2.3.2 散斑图像间互相关特性分析 ........................................................................ 22 2.4 本章小结 .................................................................................................................... 24 第三章基于激光散斑获取深度图算法 .......................................................... 25 3.1 基于散斑图的深度测量公式推导 ............................................................................ 25 3.1.1 深度测量公式推导 ........................................................................................ 25 3.1.2 散斑模式的研究 ............................................................................................ 26 3.2 深度与偏移量关系研究 ............................................................................................ 28 3.2.1 深度与偏移量的理论计算 ............................................................................ 28 3.2.2 深度与偏移量的实验测量 ............................................................................ 30 3.3 深度图获取的实现算法 ............................................................................................ 31 3.3.1 深度图获取流程 ............................................................................................ 31 3.3.2 散斑图像的预处理 ........................................................................................ 33 IV

3.3.3 快速块匹配法检测深度的过程 .................................................................... 35 3.4 本章小结 ..................................................................................................................... 38 第四章深度图获取算法的优化 ...................................................................... 39 4.1 计算速度优化 ............................................................................................................. 39 4.1.1 匹配块大小讨论 ............................................................................................ 39 4.1.2 动态搜索块 .................................................................................................... 41 4.1.3 匹配策略优化 ................................................................................................ 43 4.2 深度图精度优化 ......................................................................................................... 44 4.2.1 影响深度精度的因素 .................................................................................... 45 4.2.2 像素偏移量插值 ............................................................................................ 46 4.3 本章小结 ..................................................................................................................... 47 第五章实验与性能分析 .................................................................................. 48 5.1 实验环境 ..................................................................................................................... 48 5.2 仿真实验过程与实验结果 ......................................................................................... 48 5.3 速度与计算量分析 ..................................................................................................... 49 5.4 精度分析 ..................................................................................................................... 50 5.5 本章小结 ..................................................................................................................... 51 第六章总结与展望 .......................................................................................... 52 6.1 工作总结 ..................................................................................................................... 52 6.2 研究展望 ..................................................................................................................... 53 参考文献 ............................................................................................................. 54 攻读学位期间取得的研究成果 ......................................................................... 57 致谢 .............................................................................................................. 58 V

资料库

基于激光散斑的深度图获取算法研究.pdf

相关推荐

人工智能

热门标签

最新资料