基于可穿戴传感技术的步态参数建模与跌倒监测方法

Gait parameters modeling and fall detection method based on wearable sensing technology

在跌倒与认知功能障碍等神经疾病的临床诊断与康复治疗中,步态分析能有效的作为诊断依据,并对跌倒风险与神经疾病患者的康复情况进行评价,帮助制定康复治疗方案和评价疗效。步态数据采集是步态分析的重要环节。其中跨步长和离地高度作为步态数据的一部分,在实际医疗使用中具有较强的参考价值。现有的步态数据采集设备尽管功能强大,采集的数据种类也十分丰富,但是有着操作麻烦、价格昂贵、空间限制等缺点。简易的步态数据采集设备一般都难以采集到跨步长和离地高度等关键数据,难以满足实际需要。本文针对以上的问题,提出了一种通过三轴加速度传感器和陀螺仪采集足部角度数据并通过BP神经网络进行机器学习来推算跨步长和离地高度数据的方法。通过对训练集进行学习训练与交叉认证,得到了跨步长86%,离地高度80%的准确率,论证了通过足部跨步过程中的角度数据推算跨步长和离地高度的可行性,并且利用角度数据进行跌倒监测的初步试验,监测成功率为93. 75%。

Gait analysis has been applied as an effective indicator in clinical diagnosis and rehabilitation of neurological diseases such as falls and cognitive impairment,which could also help with making rehabilitation plan and evaluating fall risk and rehabilitation efficiency.Gait data collecting is the fundamental step for further gait analysis,and step lengths and heights are two essential gait parameters for their important clinical practice value.Current gait data collectors have multi-functions and can collect various types of data,but with shortcomings of complicated operation,high cost,large space requirements,et al.On the other hand,it is difficult for simple gait data collectors to collect some important data such as step lengths and step heights directly.To address the above problems,we proposed an approach for collecting step lengths and heights by collecting plantar angles using a three-axis gyroscope and an acceleration sensor.By using Back Propagation neural networks,we built models for calculating step lengths and heights of gait.After training and cross validation on training set,the approach reached an accuracy of 86 % during calculating step lengths and 80 % for step heights.It shows that plantar angles are potential indicators for describing step lengths and heights,and also for fall detection with detection rate of 93.75 % in the experiments.