【作者】 张天厚；

【导师】 李长春；

【作者基本信息】 济南大学 ， 机械电子工程， 2010， 硕士

【摘要】 编织袋广泛应用于大米、面粉、水泥、化肥、淀粉、饲料、棉纱、建材、化工原料、金属制品及其它产品的外包装,且这些物料的产量非常大。目前我国袋装物料的仓库搬运和装卸运输方式主要是靠肩扛和小车拉,不仅装卸工人的劳动强度大、工作效率低,而且是在粉尘的环境下工作,危害装卸工人的身心健康。采用机器人进行袋装物料的装卸,需要确定编织袋的准确位置才能装卸运输。针对编织袋装料后容易变形,出现偏离原来位置等特点,“基于机器视觉的物料袋自动识别研究”(山东省自然科学基金项目,编号:Y2007G23)就是在这样的背景下提出来的。本文以山东省自然科学基金为依托,完成了上述目标的图像处理与分析部分。本文研究内容主要包括以下几个方面:第一,设计了物料袋图像采集系统,建立了物料袋图像采集模型,采用张正友标定算法对摄像机进行了内外参数的标定。第二,对中值滤波算法进行了改进,利用改进的中值滤波算法对采集的物料袋图像进行图像去噪处理,利用直方图均衡化算法对图像进行增强,采用不同的边缘检测算子对物料袋图像进行边缘检测并对检测效果进行了分析,提出采用Sobel边缘检测算子与迭代阈值法相结合的图像分割方法,提取物料袋图像的边缘信息,得到的图像边缘清晰、连续性好,为机器人自动识别物料袋技术提供理论基础。第三,以单袋水泥袋为研究对象,采用平行双目立体视觉技术,根据视差原理,生成了水泥袋深度图像,生成的深度值即为水泥袋与摄像机基线之间的距离,通过实验数据分析得出了灰度值与深度值之间的函数关系,为机器人自动识别物料袋的三维形状提供理论基础。第四,以两袋化肥袋为研究对象,研究了物料袋码垛的图像处理与分析技术。采用改进的中值滤波算法对采集的图像进行去除噪声处理,利用Sobel边缘检测算子检测图像边缘,采用Otsu法分割图像并对图像进行二值化处理,采用数学形态学中的膨胀、腐蚀算法对图像进行处理,采用希尔迪奇(Hilditch)法对图像进行细化处理得到了两袋化肥袋的边缘特征,由于两袋化肥之间存在部分重叠,利用两袋化肥图像的面积不同,提取了上面一袋化肥的图像特征。采用区域相关匹配和直方图均衡化等算法得到了化肥袋的深度图像并提取了深度数据,可以识别出上面一袋化肥的边缘信息及其位置,为机器人自动识别物料袋提供理论基础。针对编织袋装料后容易变形,出现偏离原来位置等特点,作者采用基于机器视觉的编织袋图像处理与分析,得到的图像质量能够满足机器视觉对编织袋图像质量的要求,为机器人准确找到编织袋的位置,实现自动识别的目的打下良好的基础。更多还原

【Abstract】 Woven bags are widely used for the packaging products such as rice, flour, cement, fertilizer, starch, feedstuff, cotton yarn, architecture materials, chemical materials, metal products and so on, and the production of these materials is very great. At present, in China the method of warehouse handling and load-unloading material bags is still by workers’shoulder and wheelbarrow by pulling, not only the labor intensity of workers is very heavy and low work efficiency, but also the work place is full of dust, it is very dangerous for the workers’health. If using robot, it can load and unload material bags after recognizing the exact location. Since woven bags are easy to be out of shape after charge and remove from original position, " Research on Automatic Recognition of Material Bag Based on Machion Vision " (Natural Science Foundation of Shandong Province, Number: Y2007G23) was put forward in this context. Relying on the Natural Science Foundation of Shandong Province, this paper has done the image processing and analysis section on the above. In this paper mainly include the following aspects.Firstly, this paper has designed the material bag image acquisition system and established a machine vision measurement model, demarcated the inside and outside parameters of the camera using calibration algorithm of Zhang Zhengyou.Secondly, median filtering algorithm has been improved, the acquisitioned material bags image has been image denoising processed using the improved median filtering algorithm, enhanced the image using histogram equalization, the image edge of material bags have been detected using different edge detection operator, and the detection results have been analyzed, proposed the combination image segmentation method of Sobel edge detection operator and the iterative threshold to extract the image edge information of material bags, get a clear image edge, the edge continuity is well, provide a theoretical basis for the robot technology to the automatic recognition of material bags.Thirdly, taking a single bag cement sack as the research object, using the parallel binocular stereo vision technology, according to disparity principle, generated cement bag-depth image, the resulting value shall be the distance between the cement pocketed and the camera baseline, through the analysis of experimental data, obtained the function relationship between the gray value and the depth value, provide a theoretical basis for the robot technology to the automatic recognition of material bags.Fourthly, taking two fertilizer bags as the object of study, we study the image processing and analysis of material stacking image. Using an improved median filtering algorithm to remove noise for the acquisitioned image, using Sobel edge detection operator to detect the image edge, using Otsu method to segment image and binary the image, using the expansion, corrosion arithmetic of mathematical morphology to process the image, using Hilditch method to image processing has been exact edge features of two fertilizer bags, because there is overlap between the two fertilizer bags, utilize the difference of area of two fertilizer bags, extracted image features of a fertilizer bag above. Adopting the regional correlation matching algorithm and histogram equalization algorithm, we can get the depth image of the fertilizer bags and extract the depth data. We can recognize the fertilizer bag above and its location, it provides theoretical basis for automatical recognition material bags of the robot.Since woven bags are easy to be out of shape after charge and remove from original position, we use image processing and analysis of woven bags based on machine vision, so that it can meet the need of woven bags’ image quality for machine vision, lay a good foundation for the robot to find an accurate position of woven bags and achieving the purpose of automatic recognition.更多还原