基于计算机视觉的铝塑泡罩产品检测方法研究

【作者】 龙永红；

【导师】 谭骏珊；

【作者基本信息】 中南林学院 ， 森林工程， 2002， 硕士

【摘要】 计算机视觉在实时控制领域的应用是随着计算机硬件技术的发展和对图像处理的深入研究而逐渐发展的。本课题的研究内容是设计一套基于计算机视觉的铝塑泡罩产品包装实时检测的处理方法。以图像处理技术为基础，结合视觉技术、模式识别技术来检测出漏装、碎片、异物等不合格铝塑泡罩包装产品。 分析了利用计算机视觉进行铝塑泡罩包装产品的实时检测的硬件要求。通过现场实验确定了摄像机的要求、光源环境、图像采集卡的性能要求，设计了一套合适的计算机视觉硬件系统，并根据系统要求设计了系统软件。研究了适用于产品包装检测的图像处理算法及方案。 由于铝塑泡罩包装产品反射、折射较强，另外不合格产品中存在裂纹、发丝等异物需要识别，图像检测困难。采用线性灰度变换等方法对图像进行增强，使图像的对比度扩展。采用改进型中值滤波法和多图像平均法，在去除了噪声的基础上尽可能的保护了边缘。边缘提取是图像处理的重点和难点。采用Sobel算法对图像进行处理，使图像边缘部分突出。通过二值化将不同灰度区区分出来，这样有利于图像的特征提取。 在此基础上，利用药片的属性来进行图像分析。分别计算各位置上药片的面积A和周长值P，判断是否和理论值基本一致。然后利用C＝P~2／A来计算C是否接近于4π。通过与标准模板进行对比判断，可以识别药片是否有空泡、漏装、碎片、是否存在异物或破损等情况。根据以上图像分析结果得出产品包装是否合格。更多还原

【Abstract】 The application of computer vision in real time control develops with the development of computer hardware technology and the profound research on picture processing. The research carried out here is to design a real time detecting program based on computer vision for aluminum-plastic foamed mask packaging products. Based on the picture processing technology, the vision technology, and the pattern recognition technology, this program helps detect unqualified foamed mask packaging products made of aluminum-plastics, like those with omission assemblage, fragments, and foreign bodies.Analysis is done on the required hardware for the application of computer vision in real time detection over aluminum-plastic foamed mask packaging products. On-the-spot experiments have determined the requirements for the camera, the light source environment, and the properties of the picture collector, designed a computer vision hardware system, and the system software required for the hardware system. A picture processing algorithm and a program are worked out for package detection.The vision processing is made difficult by the comparatively strong refractiveness, reflectivity, the crazes and foreign bodies like hair existing in unqualified aluminum-plastic foamed mask packaging products. The linear darkness transform method is used to increase the contrastiveness of pictures; the improved mid-value filtering algorithm and local average method are adopted to protect the edge as much as possible on the basis of noise elimination. The edge extraction is the focus and trouble of picture processing. By using the Roberts gradient method, Laplace algorithm, and Sobel algorithm, the picture is processed and the picture edge is highlighted. To suit the characteristic extraction of pictures, the dichotomy is used to differentiate varied darkness regions.On this basis, the characteristics of tablets are used for picture analysis. The area (A) and the perimeter (P) of the tablets at each position are calculated respectively to see whether they agree with the theoretical values. Then the value of C is calculated by using the function C=P2/A to see whether it approximates 4 . Compare the value with that in the standard mother plate, unusual things like hollow foams, omission assemblage, fragments, foreign bodies or damages in the package of tablets can be detected. Results obtained from this picture analysis determine whether the product package is qualified.更多还原