“图代炮”靶面图像识别算法研究

【作者】 彭丽妮；

【导师】 徐志良； 张永；

【作者基本信息】 南京理工大学 ， 控制理论与控制工程， 2010， 硕士

【摘要】 “图代炮”系统,即通过判读瞄准镜视场图像代替实弹射击训练系统。此系统对射击目标瞬间的瞄准镜视场内图像进行处理,通过判断瞄准标志的大立标箭头与目标相对位置,判定发射炮弹是否命中目标,从而达到代替实弹射击的目的。该系统主要由三部分组成：跟踪瞄准图像摄取系统、数据采集与存储系统及图像处理与评价系统。本文主要对本系统的图像处理与评价部分进行了系统研究。图像处理包括大立标箭头的识别和目标靶的识别。在大立标箭头识别中,应用Hough变换及图像几何变换预处理图像,简要分析了各种模板匹配算法的优缺点,并提出了一种对二值图像进行模板匹配的算法,匹配速度效果良好。在目标识别方面,通过对常用图像分割算法的实验比较,选取效果较好的局部区域生长法进行目标分割。评价方面,将数据处理提升到亚像素级,通过最小二乘拟合法,使大立标箭头顶点达到亚像素级；分析了各种亚像素边缘检测方法,并根据实际图像,提出了采用角点检测和亚像素级角点定位方法判读目标靶边缘的方法。本文在VisualC++集成开发环境中基于OpenCV类库实现系统的基本功能,本课题的研究及众多实验图像的综合处理表明,本软件能够对2000米射程内的大立标箭头和目标靶进行比较准确的定位,同时满足实时性的要求,实现了自动检测与判读的功能。更多还原

【Abstract】 "Image processing supersedes artillery firing" system is a training system using image interpretation to replace artillery firing.That is to say, the system is to analyse the relatively position between the collimation sign and target firstly and then to evaluate whether the operator hits the targe. It mainly consists of image incepting subsystem of tracking and aiming, data collecting and preservating subsystem and image processing and evaluating subsystem. This paper mostly researchs the image processing and evaluating system.Image processing includes collimation sign recognition and target recognition. In collimation sign recognition,we adopt Hough transform and image geometry transform to pretreat the images, make a comparative analysis of the advantages and disadvantages among various template matching algorithms, put forward an approach of template matching with binary image and achieve relatively good results in matching pace.In target recognition, through compareing some essential image segmentation algorithms, we choose region growing effectively to recognize the target.While evaluating the result,we improve the data precision from pixel to sub-pixel. The common sub-pixel edge detection algorithms are studied, and put forward a method combining corner detecting and sub-pixel corner orientation. This method could be used to detect the edge of given images at present.This paper uses the C++ programing language and OpenCV class libraries to perform the demand of the system. Experiments show that the software can make relatively correct orientation to arrowhead and targe in 2000 meters, meet the real-time request and realize the requested function.更多还原