检索规则说明:AND代表“并且”;OR代表“或者”;NOT代表“不包含”;(注意必须大写,运算符两边需空一格)
检 索 范 例 :范例一: (K=图书馆学 OR K=情报学) AND A=范并思 范例二:J=计算机应用与软件 AND (U=C++ OR U=Basic) NOT M=Visual
作 者:叶启朗 李戴薪 南海 Ye Qilang;Li Daixin;Nan Hai(School of Computer Science&Engineering,Chongqing University of Technology,Chongqing 400054,China)
机构地区:[1]重庆理工大学计算机科学与工程学院,重庆400054
出 处:《计算机应用研究》2023年第11期3509-3514,共6页Application Research of Computers
基 金:国家自然科学基金资助项目(62141201);重庆理工大学研究生教育高质量发展行动计划资助成果(gzlcx20223194);重庆市教育委员会科学技术研究计划项目(KJQN201901133)。
摘 要:针对当前坐姿识别技术在不借助外部工具以及在动态环境下检测效果较差的情况,提出了一种基于人体骨架的坐姿识别方法,实现在任意角度环境下对坐姿的识别。人体骨架由关节和骨骼相互连接所成,其骨架信息不受动态环境和复杂背景的影响。该方法首先对任意角度下的坐姿图像进行人体姿态估计,提取出人体骨架信息,在三维空间下利用骨骼的物理连接关系构造一种用于检测坐姿的骨架图,并提出Skeleton-GCN(骨架图卷积网络)提取骨架图空间特征,将特征取平均值聚合并输入MLP层保存输出预测概率。此外,将任意角度的人体骨架坐标在三维空间下的不同角度进行投影,得到不同角度下的骨架图像,通过对比网络对骨架图像进行正面姿态估计得到正面骨架图像并输入CNN中,输出该骨架图像的预测概率,通过加权融合集成两个模态分类器预测概率并输出类别结果。最后,将所提方法应用于办公室、课堂等场景,实现了任意角度下坐姿的有效检测。Aiming at the current sitting posture recognition technology’s poor detection effect in a dynamic environment without the help of external tools,this paper put forward a sitting posture recognition method based on the human skeleton,which could recognize sitting posture at any angle environment.It represented the human skeleton by the interconnection of joints and bones,and not affected its skeleton information by a dynamic environment and complex background.The proposed method firstly performed human pose estimation on the sitting posture image at any angle to extract the human skeleton information and used the physical connection relationship of the skeleton to construct a skeleton graph for detecting the sitting posture in three-dimensional space.This paper proposed a Skeleton-GCN network to extract the spatial features of the skeleton graph,and aggregated the eigenvalues on average,then put it into the MLP layer to save the output prediction probability.In addition,it projected the human skeleton coordinates of any angle at different angles in the three-dimensional space to obtain the skeleton images at different angles.Using the contrasting network to estimate the frontal posture of the skeleton images to obtain the frontal skeleton image and put it into the CNN to output the prediction probability of the frontal skeleton image.In the final step,integrating two modal classifiers through weighted fusion to predict probability and output category results.This paper applied the proposed method to scenes such as offices and classrooms to achieve effective detection of sitting posture at any angle.
关 键 词:坐姿识别 人体姿态估计 图卷积网络 对比网络 模态融合
分 类 号:TP391.41[自动化与计算机技术—计算机应用技术]
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在链接到云南高校图书馆文献保障联盟下载...
云南高校图书馆联盟文献共享服务平台 版权所有©
您的IP:216.73.216.222