【作者】 王艳；

【导师】 孙向明；

【作者基本信息】 华中师范大学 ， 电路与系统， 2018， 硕士

【摘要】 机器视觉测量技术解决了传统测量方法不能满足社会对测量技术的要求这一问题,通过改进图像处理技术更容易提高测量系统的精度与速度,使得机器视觉测量技术逐步占据测量领域的主导地位。双目视觉是机器视觉主要表现形式,本文从基于双目视觉测量技术这一方面深入研究,并结合实验室的硅像素传感器的强子治癌束流监控器项目需求,搭建一套放疗过程中跟踪患者呼吸运动情况的系统,降低因呼吸运动使肿瘤靶区产生位移对放疗效果的影响。利用图像技术实时跟踪体表标记物运动状态反馈患者呼吸运动情况,从而推理出呼吸运动状态是否导致了肿瘤靶区产生位移。由于系统一般使用定焦相机需要进行手动调焦操作,图像清晰度一般由人工主观意识判断存在一定误差,导致系统精度不高,为提高测量系统的精度,搭建自动调焦系统自动化快速精准进行图像对焦具有实际意义。为此,本文研究了双目视觉实时跟踪系统与自动调焦系统实现的关键技术。首先,本文对系统进行了整体设计,对其中的关键技术进行了分析;然后,通过对传统的亚像素边缘检测算法进行研究与仿真,提出采用对边缘点法线方向画探测线,利用探测线上像素点之间的灰度跳变关系对边缘亚像素精确定位;接着,针对本系统使用的多个矩形标记物,结合角点特征与标记物排列特征,研究了基于多个矩形标记物特征的识别算法;随之,针对这多个矩形标记物,提出利用以每个矩形标记物的轮廓内区域作为基元进行立体匹配的匹配算法;然后,针对调焦系统,通过边缘的清晰度与图像清晰度的正比关系,提出一种优化的基于边缘特征的调焦评价函数。最后,开发了基于双目视觉实时跟踪系统与自动调焦系统。双目视觉实时跟踪系统实现图像采集、立体标定、畸变校准、滤波去噪、目标物识别、立体匹配、三维信息获取、立体显示等关键功能,可以直观获取标记物三维信息。通过单片机、电机、摄像头等部件构建自动调焦系统,实现自动化图像对焦。经过初步测试,本系统各个功能模块已经实现,亚像素边缘检测算法精度可达0.02个像素,目标物识别算法、立体匹配算法每帧处理时间分别为2.56ms与30ms。调焦评价函数对于灰度跳变较敏感、计算量小,较经典评价函数对图像清晰度变化更敏感,更快速。双目视觉呼吸跟踪系统测量误差低于±1mm,每帧图像操作耗时35ms。算法与系统实时性和精度达到预定目标。更多还原

【Abstract】 In recent years,machine vision measurement technology has solved the problem that traditional measurement methods can not satisfy the society’s requirements for measurement technology and with its features of rapidity,easy operation,no contact,high precision,and automation,and it is easier to improve the accuracy and speed of measurement systems by improving image processing techniques,making machine vision measurement technology gradually occupied the dominant position in the field of measurement.The main manifestations of machine vision technology is binocular vision,this paper conducts in-depth research from this aspects,based on binocular stereovision measurement system and the Laboratory’s project of a beam monitor using silicon pixel sensors for hadron therapy,set up a respiratory motion tracking system for tracking respiratory movements of patients during cancer radiotherapy to avoid normal target tissue damage caused by displacement of respiratory target during radiotherapy and affect radiotherapy effect.It is proposed to use real-time tracking of the movement status of the body surface marker to feedback the patient’s respiratory motion and to infer whether the state of respiratory motion leads to displacement of the tumor target.As the system generally uses a fixed focus camera,manual focusing operation is required,image clarity is generally judged by artificial subjective awareness.The accuracy of the system is not high due to certain errors due to manual operations,It is of practical significance to build an auto-focus system to automate the image focusing quickly and accurately to improve the measurement system composed of these cameras.For this reason,this paper research the key technologies for realizing the binocular vision real-time tracking system and the autofocus system.First of all,this article has carried on the overall design to the system,has carried on the analysis to the key technology among them;Then,through the research and simulation of traditional sub-pixel edge detection algorithms,draw a detection line in the direction of the edge point normal,positioning edge sub-pixels using the gray-scale relationship of pixel points on the detection line.Then,for the multiple rectangular markers used in this system,combining corner features and marker arrangement features,a recognition algorithm based on the features of multiple rectangular markers was researched.Then,for these multiple rectangular markers,proposes a matching algorithm that uses three-dimensional matching with the inner areaprimitive.Then,for focusing systems,utilizing the sharp relationship between the sharpness of edges and the sharpness of the image to put forward an optimized focus of each rectangular marker as the evaluation function based on edge features.Finally,a real-time tracking system based on binocular vision and an automatic focusing system were developed.the system of a real-time tracking system based on binocular vision,the system can accomplish the functions of image acquisition,camera stereo calibration,image radial distortion and tangential distortion calibration,median filter denoising,image target recognition,image stereo matching,target three-dimensional information acquisition and interface stereo display.The auto-focus system is constructed through components such as a single-chip microcomputer,a motor,and a camera to achieve automatic image focusing.The system has been completed.After preliminary tests,the various functional modules of the system have been implemented.the edge detection sub-pixel algorithm’s accuracy can reach 0.02 pixels;The target recognition algorithm and stereo matching algorithm have a processing time of 2.56ms and 30ms per frame,respectively;This focus evaluation functions sensitive to grayscale transitions,the algorithm is simple,and the amount of calculation is small,Better than classic evaluation function for changes in image clarity,faster A real-time tracking system based on binocular vision make the measurement error of the system lower than±1mm,andtakes 35ms to process each frame of image.The real-time and accuracy of the algorithm and system reach the predetermined goal.更多还原