基于加速度传感器的上肢运动信息采集与姿态识别

Upper Limb Motion Information Acquisition and Gesture Recognition Based on Acceleration Sensor

为了对脑卒中病人的康复训练效果进行评价,针对基于加速度传感器的人体上肢动作识别这一新兴的领域开展研究,提出了一套基于蓝牙4.0的人体上肢姿态采集系统,对患者上肢康复训练中常见的7种运动信息进行采集和姿态识别.系统包括运动信息采集、信号传输、信号去噪声、动作识别等几个主要部分.实验结果表明:将传统的时域特征和过零点特征与上四分位点和下四分位点的特征进行组合,能够更好地将曲肘侧平举与曲臂弯曲静止等动作分开,有效提高识别的准确率.与BP神经网络相比,基于径向基核函数的支持向量机(support vector madine,SVM)分类器具有明显的性能优势,获得了较好的姿态识别性能,交叉验证平均正确识别率可达90%.

In order to evaluate the effective of rehabilitation training for stroke patients, based on acceleration sensor,a new method of identifying the gesture of arm was proposed in this paper. Information collection, signal transportation, denoising and gesture recognition were concluded in our system. The information of upper limb was collected by using acceleration sensor. Wavelet transform was applied to smooth the signal in order to reduce the affection of noise. Then support vector machine （SVM） was used to distinguish seven movements by selecting an appropriate kernel function. Finally the effect of rehabilitation training was evaluated. Experimental result shows that by combining zero-crossing points,four points on the s ite,four points locus and time domain feature,bend elbow and lateral raise can be separated from other gestures. Compared with BP neural network, the SVM can achieve a good result. An accuracy of 9 0 % was reached by using the new feature and RBF kernel in our method.