多摄像头联动跟踪图像物料重识别方法  

Multi-camera linkage image tracking material re-recognition method

作　　者：陈雪文 周若萌 刘佳 江璐璐 张华强 CHEN Xue-wen;ZHOU Ruo-meng;LIU Jia;JIANG Lu-lu;ZHANG Hua-qiang(State Grid Jiangxi Procurement Co.,Ltd.,Nanchang 330077,China;State Grid Shanghai Municipal Elextric Power Company,Shanghai 201105,China;Anhui Jiyuan Software Co.,Ltd.,Hefei 230088,China)

机构地区：[1]国网江西省电力物资有限公司,南昌330077 [2]国网上海市南供电公司,上海201105 [3]安徽继远软件有限公司,合肥230088

出　　处：《信息技术》2023年第4期161-166,共6页Information Technology

摘　　要：为保证入库物资能够严格按照规定仓位进行存储,提出基于CNN的多摄像头联动跟踪图像物料重识别方法。分析膨胀、腐蚀操作过程,经过形态开、闭运算,过滤跟踪图像噪声,将阈值分割与归一化颜色模型相结合,分离背景与前景;获取不同摄像头在相同位置对应点,设置卷积层与池化层的滑动步长,探究激活函数层与全连接层特征,建立卷积神经网络模型,结合物料已知特征设置语义,引入损失函数,提取图像中物料深度信息,完成重识别。仿真实验表明,所提方法可增强图像质量,获取更多物料信息,提高识别精度。In order to ensure that the warehoused materials can be stored in strict accordance with the specified warehouse,a multi-camera linkage image tracking material re-identification method based on CNN is proposed.The expansion and corrosion operation process are analyzed,the image tracking noise is filtered through morphological opening and closing operations,and the threshold segmentation is combined with normalized color model to separate the background and foreground.After that,the corresponding points of different cameras are obtained at the same position,the sliding step of convolutional layer and pool layer are set to explore the characteristics of activation function layer and full connection layer.Then,the convolutional neural network model is established,and the loss function is introduced by the combination of known characteristics of materials.The re-recognition is completed through the extraction from the depth information of materials in the image.Simulation results show that the proposed method can enhance image quality,obtain more material information and improve recognition accuracy.

关 键 词：卷积神经网络 多摄像头联动 跟踪图像 物料重识别 形态学滤波 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]