基于深度学习的牙刷姿态识别方法  

Toothbrush gesture recognition method based on deep learning

作　　者：王化明[1] 刘茂兴[1] 熊峻峰 于金龙 WANG Huaming;LIU Maoxing;XIONG Junfeng;YU Jinlong(College of Mechanical and Electrical Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing,Jiangsu 210016,China)

机构地区：[1]南京航空航天大学机电学院,江苏南京210016

出　　处：《江苏大学学报（自然科学版）》2021年第3期298-302,共5页Journal of Jiangsu University：Natural Science Edition

摘　　要：针对牙刷分拣中的定位问题,在确定牙刷位置的基础上采用深度学习实现牙刷姿态识别.对牙刷图像进行去噪增强,通过阈值分割提取感兴趣区域,计算图像的几何矩获得牙刷的方向角和外接矩形,以外接矩形的中心作为牙刷位置.用矩形框内的牙刷图像训练残差网络模型,当模型正确率达到要求时保存该模型,用于判断图像中牙刷的姿态.测试结果表明,该方法可以快速准确地实现牙刷的位置确定与姿态识别,为机器人分拣提供牙刷位姿信息.To solve the positioning problem in toothbrush sorting,a method was proposed based on deep learning for toothbrush positioning and gesture recognition.The image of toothbrush was denoised and enhanced,and the interested region was extracted by threshold segmentation.The geometric moment of image was calculated to obtain the direction angle of toothbrush and the outer rectangle.The center of outer rectangle was taken as the center of toothbrush to determine the toothbrush position.The residual network model was trained with the toothbrush image in the rectangular box.When the accuracy of the model reached the requirement,the model was saved to judge the attitude of toothbrush in the image.The test results show that the method can quickly and accurately realize the position determination and gesture recognition of toothbrush,and provide the posture information for the robot operation.

关 键 词：牙刷分拣 姿态识别 几何矩 深度学习 残差网络模型 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]