一种基于智能手机终端的远程人体姿态监测系统

Remote human body posture monitoring system based on smart phone terminal

为了利用便携式设备准确监测老年人的跌倒状况,提出了一种基于softmax回归的多种行为模式分类识别方法,设计并实现了基于智能手机终端的远程人体姿态监测系统。首先,构建softmax分类器分析8种日常行为模式下的加速度模值特征,由于跑步时加速度模值与突然跌倒时类似,引入倾斜角特征进行二次判别,从而识别出突然跌倒行为;针对缓慢跌倒行为下加速度模值特征不明显的问题,在softmax分类器中引入躺倒时间特征,通过设置躺倒时间阈值并判断原始位置是否在时间阈值内恢复,从而识别出缓慢跌倒行为。实验与测试结果表明,该系统准确度为95.40%,特异度为95.33%,灵敏度为95.50%,具有较高的跌倒行为识别精度,对老年人的健康状态监测提供了一种可行方案。

In order to use the portable device to accurately monitor the fall of the elderly, a method of classifying and identifying multiple behavior patterns based on softmax regression is proposed, and a remote human posture monitoring system based on smart phone terminal is implemented. First, a softmax classifier is constructed to analyze the acceleration modulus characteristics under 8 daily behavior modes. Since the acceleration modulus during running is similar to that during a sudden fall, the inclination angle feature is introduced for secondary discrimination to identify the sudden fall behavior. Aiming at the problem that the acceleration modulus value characteristics are not obvious under the slow fall behavior, the lying-down time feature is introduced in the softmax classifier. The slow fall behavior is identified by setting the time threshold and judging whether the original position is recovered within the time threshold. Experiment and test results show that the accuracy, specificity and sensitivity of the system are 95.40%, 95.33%, and 95.50% respectively with high recognition accuracy for falling down behavior, which provides a feasible solution for the elderly’s health monitoring.