基于自回归模型和神经网络的跌倒检测算法

Fall detection using autoregressive model and artificial neural nets

为提高老人跌倒检测的准确性,提出基于自回归模型系数和神经网络的跌倒检测算法。采用陀螺仪获取人体运动的三维加速度和角速度值,研究人体日常活动和跌倒的运动特征,构建关于自回归系数、SMA和倾角的组合特征向量,建立神经网络分类器,实现对跌倒和日常动作的识别。实验数据共390例样本,包括9类日常动作和4类跌倒,其中195例为训练数据,另外195例为测试数据,测试结果表明,敏感度为97.2%,特异性为99.74%,准确率为98.97%,该方法达到了良好的分类性能。

A fall detection algorithm using augmented autoregressive(AR)model coefficients and artificial neural nets was proposed to improve the accuracy of the fall detection in the elderly.The activities of daily life(ADL)and falls signal were measured and analyzed using a gyroscope sensor,an augmented vector was constructed based on AR coefficients,signal magnitude areas(SMA)and title angels,and the neural network classifier was established.The falls and ADLs were recognized.A date-set recorded from young healthy volunteers performing 390 samples,including 9 kinds of ADLs and 4 kinds of falls,of which 195 cases from training set,and the other 195 cases from testing set.The results indicate that the algorithm is efficient,with sensitivity of 97.2%,specificity of 99.74%and accuracy of 98.97%.