【作者】 焦靖华；

【导师】 赵永平；

【作者基本信息】 哈尔滨工业大学 ， 仪器科学与技术， 2010， 硕士

【摘要】 汽车组合仪表是汽车中不可缺少的组成部分,负责记录和显示汽车的各种数据。根据汽车领域对汽车组合仪表美观、多功能、轻型化的要求,汽车仪表生产和检测的自动化程度需要大大提高。目前在汽车仪表检测中通常采用人工观测各表针与刻度之间的压线情况、报警灯点亮情况、以及是否有外观缺陷来检验仪表,判断产品质量是否合格。这种方法受主观因素、观测距离及疲劳程度影响严重,容易产生误差,可靠性差。数字图像处理技术的快速发展,使得机器视觉技术开始被应用于汽车组合仪表的检测中。本课题旨在基于机器视觉技术,采用图像处理的方法实现汽车组合仪表读数的自动识别,代替人工识别这一枯燥劳累、可靠性差的方法。本文首先介绍了机器视觉及汽车组合仪表的概念和发展应用现状。接下来阐述了图像采集和图像处理的总体方案。然后详细介绍了表盘读数识别的图像处理理论和方法,研究了图像预处理的方法,通过灰度化、滤波、二值化,使图像质量有了很大提高;提出区域标记和基于指针区域面积、形状、颜色特征提取指针区域,并取得良好效果;采用Hough变换和最小二乘法实现直线提取;采用角度法和数字法两种确定读数的方法,并提出以过线数为特征进行二级数字识别,快速有效;将组合仪表排列方式分为相连和独立两类,提出不同的分割方法将组合仪表图像分割为单个表盘的分割方法。本文在详细的分析图像处理算法基础上,在MATLAB平台上编制了汽车组合仪表读数识别软件,并通过实验验证该软件的速度和准确性,并分析了误差的来源,准确性可基本满足实际要求,速度有待提高,具有一定的应用价值。更多还原

【Abstract】 Automobile Instrument is an indispensable vehicle component, which is responsible for recording and displaying the various data. According to requirements of more beautiful, more functional, and lighter for the Automobile Instrument, manufacturing and testing of Automobile Instrument is supposed to be substantially improved. Now in automotive instrument testing area manpower methods of judging whether the gauge indicators are pressing or between graduation lines, whether the warning lights are lit, whether there are surface defects are commonly used to determine the quality of products. These methods are influenced by subjective factors, detecting range and fatigue degree, which would lead to big errors and poor reliability. As the rapid development of digital image processing technology, computer vision is beginning to be used in Automobile Instrument detection.Aim of the subject is that based on machine vision technology, image processing methods could be used in automatic identification of Automobile Instrument readings, instead of manually identify which is boring and tired with poor reliability. Firstly, this paper introduces the concepts of machine vision and automobile instrument, development and application in related area. Secondly, overall program of image capture and image processing is described. Thirdly, image processing theory and methods are introduced in detail. Methods of image preprocessing are studied, the image quality is greatly improved by gray level transformation, filtering, binarization etc; region labeling and indicator-region-extraction methods based on size of the area, shape and color feature are propose, and achieved good results; line extraction is achieved by Hough Transform and Least Square Method; the reading is determined in two ways of angle method and number method, and line-cross-method is proposed to identify numbers quickly and effectively; the automobile instrument arrangement is divided into two kinds: linked and independent, different segmentation methods are proposed to divide automobile instrument into single segmentation.In this paper, based on detailed analysis of image processing algorithm, recognition software for automobile instrument reading is developed on MATLAB platform, the speed and accuracy is verified by experiments,the sources of errors are analyzed. The acuracy can satisfy the practical requirements ,however the speed is to be improved, which is of certain practical value.更多还原