基于加速度信号多特征踝关节动作模式识别

Ankle Motion Pattern Recognition of Multi-Feature Based on Acceleration Signal

人体动作识别技术在虚拟现实、机器人、体感游戏等诸多领域具有较大的应用潜力.为了有效识别下肢踝关节的不同动作模式,首先通过DELSYS信号采集仪,获取踝关节执行趾屈、背伸、内翻、外翻4种动作模式时的三轴加速度信号,利用小波去噪滤除信号采集过程中的干扰和噪声;然后,提取踝关节三轴加速度信号的绝对积分平均值、方差、两轴之间的相关系数以及幅度峰值和幅度均值5种特征参数,并融合组成特征向量输入支持向量机分类器进行动作模式识别.实验结果表明:将加速度信号在特征层上融合再进行踝关节动作识别,每种动作的平均分类正确率均可达到90%以上,该研究方法可以应用于虚拟现实游戏及康复机器人等领域.

Human body motion recognition technology can show great potential for application in many fields such as virtual reality,robotics and somatosensory games.The purpose is to effectively identify different action modes of lower limb ankle.First of all,through the DELSYS signal acquisition instrument,we can obtain four kinds of triaxial acceleration signal of the ankle which is toe flexion,dorsiflexion,foot varus and foot eversion.And we can use the method of wavelet denoising to filter interference and noise which is generated during the signal acquisition.Then,five characteristic parameters of ankle three axis acceleration signals are extracted,including absolute integral average,variance,the correlation coefficient between two axes,amplitude peak and mean amplitude.Next,in order to implement the action pattern classification,the feature vector combined with five characteristic parameters are input to support vector machine.The experimental results show that the average classification accuracy of each action can reach more than 90%,when the acceleration signal features are fused to recognize the ankle movements.This research method can be applied in virtual reality games and rehabilitation robots.