基于Logistic回归的零件图像区域提取  

Extraction of parts region based on Logistic regression

作　　者：赵鹏[1] 李大寨[1] 王韬[1] 

机构地区：[1]北京航空航天大学机械工程及自动化学院,北京100191

出　　处：《计算机应用研究》2017年第4期1265-1268,共4页Application Research of Computers

摘　　要：针对机器人自动化生产线中视觉定位零件的问题进行了研究,为了能够快速地在复杂背景中提取出零件图像区域,提出了一种基于Logistic回归的零件图像区域提取的机器视觉算法。首先,在工作场景图像中采集若干正类和负类样本点;然后,将样本点的BGR值进行格式上的整理并保存;其次,将整理的数据代入按照梯度上升法推导的公式中,计算Logistic最佳回归系数;最后,载入待处理图像,使用Logistic回归分类器对其中的每个像素进行分类,提取出零件区域。实验表明,该算法提取出的零件图像区域边缘清晰,实时性高,且由于输入为像素点的BGR值,所以对零件的外形没有要求,也不需要额外的夹具。This paper proposed an algorithm based on Logistic regression for extacting component region in robot automatic production line. Firstly, it collected a number of positive and negative samples in the image of the work scene. Secondly, it formatted and saved the BGR value of sample points. And then, it calculated the Logistic optimal regression weight, according to gradient ascent method. Finally, it extracted region of component with Logistic regression in new image. Experimental resuits show that the proposed algorithm is well and real-time. Because input is BGR value of pixel, the shape of the parts is not restricted, and does not require additional fixture.

关 键 词：零件图像 区域提取 机器视觉 LOGISTIC回归 梯度上升法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]