摘要
针对人体多运动模式识别中识别精度低的问题,提出一种基于MEMS惯性传感器的人体多运动模式识别算法。该算法选取MEMS加速度传感器的时域特征作为模式识别特征量,提取MEMS角速度传感器的时域特征作为二次识别的特征量,能够准确识别走、跑、站立、上楼、下楼、趴倒、躺倒、倒退多种运动模式。识别过程采用分层识别算法,同时使用支持向量机识别区分难度较大的两类行为动作。嵌入式消防单兵定位系统平台的实验验证表明,利用该算法能够识别消防单兵的多运动模式,平均识别精度达到94%以上。
A recognition algorithm for multiple human motion patterns based on MEMS inertia sensors was proposed for the problem of low accuracy of pattern recognition in recognizing multiple human motion patterns. Time domain features of a MEMS acceleration sensor were selected as the features of pattern recognition. Time domain features of the MEMS gyroscope were adopted as the features of secondary recognition. The algorithm can accurately identify various movement patterns, including walking, running, standing, upstairs, downstairs, forward falls, backward falls and backward walking. A layering recognition algorithm was used in the recognition process. A support vector machine was trained to recognize such two types of motion patterns as they were difficult to distinguish. The experiments based on the platform of the embedded fire soldier positioning system show that the recognition algorithm can recognize multiple motion patterns of fire soldiers, and the average accuracy is above 94%.
出处
《中国惯性技术学报》
EI
CSCD
北大核心
2016年第5期589-594,共6页
Journal of Chinese Inertial Technology
基金
国家自然科学基金资助项目(51175535)
国际联合研究中心科技平台与基地建设(cstc2014gjhz0038)
重庆市基础与前沿研究计划项目(cstc2015jcyj BX0068)
重庆邮电大学博士启动基金(A2015-40)
重庆邮电大学自然科学基金(A2015-49)
关键词
MEMS惯性传感器
运动模式
识别算法
时域特征
MEMS inertia sensors
motion pattern
recognition algorithm
time domain features