基于梯度的曲面图像匹配方法研究

CORC > 沈阳自动化研究所 > 中国科学院沈阳自动化研究所 > 光电信息技术研究室

题名	基于梯度的曲面图像匹配方法研究
作者	孙万明
学位类别	硕士
答辩日期	2016-05-25
授予单位	中国科学院沈阳自动化研究所
导师	吴清潇
关键词	曲面目标匹配梯度并行计算模板匹配
其他题名	Object with Curved Surface Matching Method Based on Gradient
学位专业	控制工程
中文摘要	机器视觉是机器人应用研究中的一个主要研究方向，广泛应用于目标识别与定位。模板匹配是机器视觉中常用的目标识别定位方法，本文在现有的模板匹配算法的基础上，从曲面在平行光照下的成像特性出发，提出了一种针对图像中带有曲面的目标的匹配方法。曲面在数字图像中的成像具有鲜明的特点，即曲面上的不同位置由于其法线方向不同，对光线的漫反射方向不同，使得曲面在灰度图像中表现为连续变化的灰阶分布。本文选择用灰度梯度作为描述子来描述曲面表面的灰度变化情况，曲面上一点的灰度梯度方向指向该点邻域内灰度增长最快的方向，梯度强度为梯度方向上灰度增长的程度。图像预处理阶段，根据模板图像建立不同旋转角度的匹配模板，计算所有模板的梯度描述。匹配过程中，利用图像金字塔和OpenMP、C++ AMP技术在CPU和GPU上进行并行计算，对匹配过程进行提速。利用归一化相关系数法计算模板与目标的曲面间的相似性，最终给出目标在图像中的位置和旋转信息。研究过程中，建立了由相机、光源和计算机组成的硬件实验平台，并编写匹配定位实验软件。经过大量的匹配实验和理论分析，验证了本文的方法能够对包含曲面的目标的精确匹配和定位，并且满足光照强度变化不敏感以及小范围内的尺度变化和光照方向变化的鲁棒性，具有很好的稳定性和实用性。
英文摘要	Machine vision is an important research direction of robotic application research, which is widely applicated in object recognition and locating. Template matching is frequently used in machine vision as object recognition and locating method. A new template matching method that aimed at curved surface in digital image, is proposed in this paper which is based on common template matching method. Imaging of curved surface in digital image is pretty special that curved surface is present as continuous changing of gray value in gray level image. While different position on curved surface has different normal orientation and its diffuse reflection direct to different orientation. In my method, gradient of gray level pixel is used as the descriptor of gray level changing of curved surface. A gradient vector directs to the orientation that gray value of pixel increased most, and vector module indicates the increasement. Reference to the scale of template image, a template pyramid will be built before matching. The template pyramid is consist of gradient vectors arrays in different rotation angle and different scale. A searching strategy based on template pyramid and CPU and GPU parallel computing technique named OpenMP and C++ AMP will be used in matching process in order to accelerate the matching speed. Using normalized inner product method as the similarity measurement function for template and object matching. The location and rotation information of object will be outputted if matched. A hardware experiment platform consist of camera, light source and computer and curved surface matching software were built during the researching period. It was tested that my method has the capacity of accurate match and locate objects with curved surface after a lot of matching experiments and theoretical analysis. And tested that the curved surface matching method is not sensitive to illumination intensity changing and robust to scale change and illumination direction in small scale, which make the method stable and practical.
语种	中文
产权排序	1
页码	55页
内容类型	学位论文
源URL	[http://ir.sia.cn/handle/173321/19625]
专题	沈阳自动化研究所_光电信息技术研究室
推荐引用方式 GB/T 7714	孙万明. 基于梯度的曲面图像匹配方法研究[D]. 中国科学院沈阳自动化研究所. 2016.