【作者】 陈伟；

【导师】 方康玲；

【作者基本信息】 武汉科技大学 ， 控制理论与控制工程， 2008， 博士

【摘要】 随着信息技术的发展,机器视觉技术作为一种新的监控技术在工业生产领域得到广泛的应用,特别是对于高温、粉尘、振动、强电磁等复杂工业环境下运动目标的检测和定位问题,机器视觉技术提供了一种低成本和高性能兼顾的解决方案。由于工业现场环境恶劣,存在振动、粉尘等干扰因素,且有些工业现场无法安装固定的视觉系统照明光源,这些干扰因素会对CCD摄像机的成像质量带来不利影响。此外,工业生产控制对机器视觉技术检测的实时性和准确性都有较高的要求,因此复杂工业现场下快速、精确的视觉目标检测方法成为目前机器视觉应用研究的热点和难点问题之一。本文以钢坯加热炉内钢坯定位控制为研究背景,对复杂工业现场下运动目标检测方法进行了研究。论文的主要研究内容如下:Otsu方法是一种经典的非参数、无监督自适应阈值选取方法,在灰度图像分割上得到广泛的应用,针对Otus方法在多阈值图像分割中阈值寻优计算量大的问题,提出了一种基于粒子群的Otsu多阈值混合优化算法,图像分割实验表明该混合优化方法可以减少Otsu方法在多阈值图像分割中最优阈值的寻优时间。在分析标准粒子群算法的基础上,提出了一种基于局部搜索算子的改进粒子群算法,数值仿真实验结果表明局部搜索算子可以改善粒子群算法的性能。将改进粒子群算法与FCM算法结合,提出了一种基于改进粒子群算法的混合FCM优化算法,图像聚类分割实验表明该混合优化算法可以改善FCM图像聚类分割的效果。由于工业现场环境比较恶劣,视觉系统监控场景的光线会受工业现场条件变化的影响,背景光线的变化会导致图像背景区域内像素点灰度值的变化,如果未能检测出背景光线的变化,会造成视觉系统检测到错误的目标运动信息。为了实时检测出背景光线的变化,提出了一种简便的光线检测方法,通过比较图像局部特征区域的灰度均值变化,检测背景光线的变化情况。对有些复杂工业现场存在的无法安装固定视觉系统光源,背景光线会随工业现场条件的变化而变化的问题进行了研究,提出了一种固定场景下的PSO多背景图像建模方法,从背景图像序列中提取多个比较有代表性的背景图像建立背景图像模型,通过背景光线变化检测方法和PSO背景图像匹配算法,从多背景图像中选取与当前输入图像最相关的背景图像作为当前背景图像。钢坯加热炉是轧钢作业在轧制前重要的热加工设备,钢坯定位的精度对加热炉的运行至关重要。针对视觉钢坯定位控制系统存在的钢坯定位精度低、定位偏差波动较大等问题,将PSO多背景图像检测方法嵌入视觉钢坯定位控制系统中,通过PSO多背景图像检测方法对加热炉内钢坯运动边缘进行检测,钢坯定位的测量数据表明PSO多背景图像检测方法可以有效提高视觉钢坯定位控制系统的钢坯定位精度。更多还原

【Abstract】 With the development of information technology, machine vision, as a new detecting and control method, has been applied in many industrial fields. Machine vision is the best trade-off between high accuracy control and low cost under complex industrial environments such as high temperature, dust, harsh EMI, and so forth.Vibration, dust and no fixed lighting would lead the CCD image degradation in harsh industrial conditions. The machine vision system under industrial environments should detect the objects accurately and rapidly. Therefore, it is one of the hot issues to develop a fast, precision detecting algorithm based on vision under complex industrial conditions.In this dissertation, the research is focused on the object detecting problems under complex industrial environments, and the billet detecting and locating control in kiln is as a background research project. The main works are as follows:The Otsu method is one of very efficient thresholding method for the gray image. However, its computation would become more complex in the case of multilevel thresholding. A multilevel thresholding algorithm that is based on particle swarm optimization is developed to solve the complexity of the Otsu method in multilevel thresholding. The experimental results show that the PSO-Otsu can provide better effectiveness on experiments of image segmentation.In order to improve the performance of PSO, an improved PSO using a local searching operator is proposed. The several benchmark functions are used to testify, and the results show the improved PSO has a better performance. A FCM image segmentation based on the improved PSO is proposed to improve the performance of FCM. The experimental results show that the hybrid optimization scheme can provide better effectiveness for image segmentation.The background illumination would change under harsh industrial conditions, and lead to the gray change of pixel in a background image region. The vision system would extract incorrect object information if the illumination changes are not detected. A simple detecting method is proposed to detect the background illumination changes by detecting the change of mean gray value in the local characteristic regions between two sensed images.There is not a fixed lighting in some complex industrial conditions, and the illumination changes would change with the industrial conditions changes. A multi background images model is proposed to adapt to the illumination changes under complex industrial environments. The multiple background images model is extracts from the background video in the processing of off-line learning, and swaps the background image, which is the most similar image to the current sensed image in the multi-background images, by detecting the illumination changes and matching background images using PSO.The heating kiln plays an important role in steel milling in the Steel Mill. The vision billet location control system has some problems in the billet location control, such as the low location precision, the high variation of location deviation, and so on. The multi background images detecting method using PSO is applied in the location control of billets. The result shows PSO multi background images detecting method can improve the precision of the vision billet location control system.更多还原