基于改进的YOLOv3的托盘检测方法研究  

Research on Pallet Detection Method Based on Improved YOLOv3

作　　者：王平凯[1] 孙光泽 朱芮萱 Wang Pingkai;Sun Guangze;Zhu Ruixuan(School of Electrical and Mechanical Engineering,Changchun University of Technology,Changchun 130012,China)

机构地区：[1]长春工业大学机电工程学院,长春130012

出　　处：《机电工程技术》2021年第7期29-32,共4页Mechanical & Electrical Engineering Technology

基　　金：吉林省发改委项目(编号:2020C018-3)。

摘　　要：针对仓储环境下叉车机器人托盘识别的应用场景,以及提高托盘目标检测的准确性和鲁棒性,提出了一种基于YOLOv3算法改进后的物体识别方法。运用K-Means++聚类方法重新聚类出更适合托盘检测的Anchor Box,通过分析托盘成像在图像坐标系中横轴和纵轴的密度分布,继而调整了划分网格机制,改进损失函数。并在运用数据增强手段的托盘数据集上进行训练以及测试,与其他算法进行对比,结果显示基于改进的YOLOv3托盘检测方法在测试集上的准确率达到94.6%,识别速率达到47帧/s。Aiming at the application scenario of forklift robot pallet recognition in storage environment,and to improve the accuracy and robustness of pallet target detection,an object recognition method based on the improved YOLOv3 algorithm was proposed.The K-means++clustering method was used to re-cluster the Anchor Box that was more suitable for pallet detection.By analyzing the density distribution of the horizontal axis and the vertical axis of pallet imaging in the image coordinate system,the grid partition mechanism was adjusted and the loss function was improved.Training and testing were carried out on the pallet data set by using data enhancement methods and compared with other algorithms,the results show that the improved YOLOv3 pallet detection method has an accuracy rate of 94.6%and a recognition rate of 47 fps on the test set.

关 键 词：YOLOv3算法 目标检测 仓储环境 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]