基于机器视觉罐装食品瓶盖字符识别与缺陷检测

发布时间：2018-05-07 16:37

本文选题：柱面校正 + 三维标定　；参考：《天津科技大学》2016年硕士论文

【摘要】：随着经济的发展,人们生活水平不断提高,食品的品质和安全已经成为社会关注的焦点。人们对食品生产的要求也变的越来越高,生产日期是我们评估食品安全与否的一个重要标准,因此,保证生产日期等相关信息正确清晰的标注是食品生产过程中的一个重要环节。利用机器视觉技术进行相关信息标注的缺陷检测具有非常广阔的市场需求。本文对于打印在瓶体/瓶盖表面的字符信息的缺陷检测方法进行了研究。由于瓶体的形状,表面打印的字符和标签不可避免的会发生变形,这些变形会对后续的图像分割、字符识别和缺陷检测等环节造成很大的困难。研究了基于三维标定的柱面图像校正算法,该方法利用三维标定准确计算出真实场景中瓶体相对相机的位姿,然后借助标定的位姿信息还原出瓶体的三维模型,建立起瓶体表面点与图像像素点之间的联系,从而将柱面展开为平面。通过图像的校正后,瓶体表面的标签和字符都被展开标准图像,大大降低了识别和缺陷检测的难度。字符识别与检测的具体步骤为:首先对展开后的字符图像进行字符分割,然后提取字符的一些特征组合来完成多层感知器识别模型的训练对字符进行识别。最后利用模板对比的方法将合格字符图像模板与待检字符模板进行详细对比完成字符缺陷的检测。本文还设计了一套自动光学识别和检测系统,根据图像校正后的结果将不同角度拍摄的瓶体图像拼接出360度全景图像,探索了瓶体表面标签任意位置信息的识别和检测,大大扩展了系统的应用范围。
[Abstract]:With the development of economy and the improvement of people's living standard, the quality and safety of food have become the focus of attention. People's requirements for food production are also becoming higher and higher. Production dates are an important criterion for our assessment of food safety, so, It is an important step in the food production process to ensure that the relevant information such as production date is correctly and clearly marked. There is a broad market demand for defect detection of information tagging using machine vision technology. In this paper, the defect detection method of character information printed on the bottle / bottle cap surface is studied. Because of the shape of the bottle, the characters and labels printed on the surface will inevitably deform, which will cause great difficulties to the subsequent image segmentation, character recognition and defect detection. The cylindrical image correction algorithm based on 3D calibration is studied. The method uses 3D calibration to accurately calculate the position of the bottle body relative to the camera in the real scene, and then restores the three-dimensional model of the bottle body by using the calibrated position and pose information. The relationship between the surface points of the bottle and the pixel points of the image is established, and the cylinder is expanded into a plane. After image correction, the labels and characters on the bottle surface are expanded into standard images, which greatly reduces the difficulty of recognition and defect detection. The specific steps of character recognition and detection are as follows: first, character segmentation is carried out on the expanded character image, and then some feature combinations of characters are extracted to complete the training of multi-layer perceptron recognition model for character recognition. Finally, the qualified character image template and the character template to be checked are compared to complete the character defect detection. An automatic optical recognition and detection system is also designed in this paper. According to the results of image correction, 360 degree panoramic images are stitched from different angles, and the recognition and detection of arbitrary position information of bottle surface labels are explored. The application range of the system is greatly expanded.
【学位授予单位】：天津科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TS297;TP391.41

【相似文献】