[1]陈浩龙.基于卷积神经网络的多传感器下坐姿识别研究[J].计算机技术与发展,2021,31(11):183-188.[doi:10. 3969 / j. issn. 1673-629X. 2021. 11. 030]
　CHEN Hao-long.Research on Sitting Posture Recognition Based on Multiple Sensors and Convolutional Neural Network[J].,2021,31(11):183-188.[doi:10. 3969 / j. issn. 1673-629X. 2021. 11. 030]
点击复制

基于卷积神经网络的多传感器下坐姿识别研究()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
31
期数:
2021年11期
页码:
183-188
栏目:
应用前沿与综合
出版日期:
2021-11-10

文章信息/Info

Title:
Research on Sitting Posture Recognition Based on Multiple Sensors and Convolutional Neural Network
文章编号:
1673-629X(2021)11-0183-06
作者:
陈浩龙
南京中医药大学 人工智能与信息技术学院,江苏 南京 210046
Author(s):
CHEN Hao-long
School of Artificial Intelligence and Information Technology,Nanjing University of Chinese Medicine,Nanjing 210046,China
关键词:
亚健康疾病预防; 坐姿识别; 多传感器; 数据预处理; 卷积神经网络
Keywords:
sub-health disease prevention; sitting posture recognition; multiple sensors; data preprocess; convolutional neural network
分类号:
TP391
DOI:
10. 3969 / j. issn. 1673-629X. 2021. 11. 030
摘要:
亚健康疾病困扰着当代人群,不良坐姿是颈椎病、腰椎间盘突出等疾病的最重要形成因素。 为了在纠正不良坐姿,培养良好坐姿习惯上能有更为人性化的反馈,需要对颈部、背部、腰部进行监测来识别不同坐姿。 文中将在坐姿识别上提出一个较为全面的解决方案。 相较基于视频图像识别的坐姿识别方案,提出利用多传感器采集数据,通过神经网络进行坐姿识别的方案可以避免环境因素的影响,更加精确地监测坐姿;比起单个传感器,由多个九轴传感器联合的数据采集系统,从颈部、背部、腰部,更为全面地采集人体上身部位活动数据;卷积神经网络在特征提取工程上收效显著,能更快、更准确地对不同坐姿进行多分类处理,将获取的数据进行预处理以构建数据集及标签之后输入至卷积神经网络进行深度学习。 实验最终的平均准确率在 96. 78% 以上,可以满足坐姿快速精确识别的需求。 该方案具有布置灵活、监测面广、反馈迅速、精准度高、成本低等优点。
Abstract:
The sub-health diseases are perplexing the contemporary population,and the bad sitting posture is the most important factor in the formation of cervical spondylosis,lumbar disc herniation and other diseases. In order to have more humanized feedback on correcting bad sitting posture and cultivating good sitting habits,we need to monitor the neck,back and waist to identify different sitting positions.In this paper,we will propose a more comprehensive solution on the recognition of sitting posture.Compared with the human activity recognition based on video and image recognition, the scheme to recognize human activity based on multi-sensor data acquisition and neural network is proposed, which can avoid the influence of environmental factors and monitor sitting posture more accurately. The data acquisition system composed of multiple nine axis sensors in series can collect the activity data of upper part of human body more comprehensively from neck,back and waist. Convolution neural network is effective in feature extraction engineering, and it can classify different sitting posture more quickly and accurately. In this paper,the acquired data are pre-processed to construct data sets and labels,and then input to convolution neural network for deep learning. The final accuracy of the experiment can reach 96. 78% ,which can meet the needs of rapid and accurate recognition of sitting posture. The scheme has the advantages of flexible layout,wide monitoring range,rapid feedback,high accuracy and low cost.
更新日期/Last Update: 2021-11-10