基于特征重构的类圆形密集嘴棒精确计数  

Accurate counting of crowded circle-like cigarette filters based on feature reconstruction

作　　者：钱基德 钱基业 杜冬[1] 刘长华[1] QIAN Jide;QIAN Jiye;DU Dong;LIU Changhua(Key Laboratory of Flight Techniques and Flight Safety,Civil Aviation Flight University of China,Guanghan Sichuan 618307,China;Electric Power Research Institute,State Grid Chongqing Electric Power Corporation,Chongqing 401123,China)

机构地区：[1]中国民用航空飞行学院民航飞行技术与飞行安全重点实验室,四川广汉618307 [2]国网重庆市电力公司电力科学研究院,重庆401123

出　　处：《计算机应用》2020年第S02期147-150,共4页journal of Computer Applications

基　　金：中国民用航空飞行学院面上项目(J2018-58);2019民航局教育类项目。

摘　　要：类圆形颗粒计数问题是工业生产过程中提高自动化水平的一个瓶颈,开发准确率高、实时性高、鲁棒性好的自动在线计数设备具有重大意义。以嘴棒计数为研究对象,基于特征重构的算法,实现在线实时非接触式的密集嘴棒精确计数。通过面阵相机采集已装箱的密集棒材俯视图作为计数对象,首先通过形态学算法提取棒材角点信息;其次结合棒材尺度信息,并采用归一化能量特征重构棒材区域特征;然后采用最大稳定极值区域算法准确分割出孤立的棒材区域,再根据该区域实现了棒材精确计数。该方法能有效应对密集棒材图像的亮度不均,对棒材形变鲁棒性好。实际测试验证了所提方法的可行性,在50 ms时间内能精确计数图像中的数千只棒材,平均计数准确率可达99.95%。The counting problem of the circle-like granules is a bottleneck to improve the automation level in the industrial production process.It has great significance to develop automatic online counting equipment with high accuracy,high real-time,and good robustness.Taking the cigarette filters counting as the research object,the algorithm based on feature reconstruction achieved a real-time non-contact crowded cigarette filters precise counting.Using the top view of the crowded cigarette-filters captureed by a plane array camera as a counting object,the morphological algorithm was firstly used to extract the information of the corner of the cigarette filters.Then,the corner feature information of the cigarette filters was reconstructed by normalized energy feature combined with scale information.Then,Maximum Stable Extremum Region(MSER)algorithm was used to segment the isolated cigarette filters region.Finally,precise filters counting was achieved according to the region.The proposed algorithm can effectively deal with the uneven brightness of crowded cigarette filters images,and has good robustness to cigarette filters deformation.The actual field test verifies the feasibility of this method,a frame image of thousands of the cigarette filters can be counted in 50 ms,and the average count accuracy is up to 99.95%.

关 键 词：类圆形颗粒 计算机视觉 低对比度 特征重构 最大稳定极值区域 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]