深度学习的人体图像半自动标注系统  

Deep Learning Based Semi-Automatic Labeling System for Human Images

作　　者：高慧 张继威 来扬 王文东[3] GAO Hui;ZHANG Ji-wei;LAI Yang;WANG Wen-dong(School of Computer Science(National Pilot Software Engineering School),Beijing University of Posts and Telecommunications,Beijing 100876,China;Systems Engineering Research Institute of China State Shipbuilding Corporation,Beijing 100036,China;State Key Laboratory of Networking and Switching Technology,Beijing University of Posts and Telecommunications,Beijing 100876,China)

机构地区：[1]北京邮电大学计算机学院(国家示范性软件学院),北京100876 [2]中国船舶工业系统工程研究院,北京100036 [3]北京邮电大学网络与交换技术国家重点实验室,北京100876

出　　处：《北京邮电大学学报》2021年第1期104-109,共6页Journal of Beijing University of Posts and Telecommunications

基　　金：中央高校基本科研业务费提升科技创新能力行动计划项目(2019XD-A12);北京市自然科学基金-海淀原始创新联合基金项目(L182034)。

摘　　要：针对目前数据标注过于依赖硬件、手动数据标注效率低下的问题,提出了基于深度学习的人体图像半自动标注系统.系统通过对算法进行改进,增加人体关键点个数进行特征提取和加入运动信息的约束,提高了视频分阶段标注的准确率.使用真实数据集仿真实验证明了通过深度学习算法进行数据标注的可行性,并且使用半自动标注的速度快、准确率高.In view of the problem that data labeling is too dependent on hardware and manual data labeling is inefficient,a semi-automatic labeling system for human images based on deep learning is proposed.By improving the algorithm,the system increases the number of key points of the human body for feature extraction and adds motion information constraints,which improves the accuracy of video staged annotation. Experiments that employs real data sets prove the feasibility of data labeling by deep learning algorithm,and using deep learning algorithms for semi-automatic labeling is faster and more accurate.

关 键 词：图像标注 半自动 深度学习 人体重建 

分 类 号：TP29[自动化与计算机技术—检测技术与自动化装置]