基于二值传感网络及隐私保护的人物室内动态定位、多行为识别与老人摔倒实时监测方法研究

In this study, we use the binary data of activities collected by sensor network to realize more complex human behavior analysis (HBA). We propose the optimization method of binary activity data, and the applied theory of sensor network to top-view camera. Comparing with traditional digital image/video processing method, our method and theory solve the problems of the large amount of data processing, the complexity of processing algorithm and the difficulty of realizing real-time data processing, make it possible to use small amount of data in indoor environment to complete more complex missions. In previous research, by modeling the motion of moving object, based on the least error criterion function principle and extreme iteration method, the dynamic localization method was proposed. The localization accuracy was close to 0.3 m (the size of environment was 15.0 m * 8.5 m). Based on the proposed optimization method of the binary activity data, the distribution of pixel and the intensity of activities is more distinctive. The effect of multi-activities recognition will be improved significantly. The detect rate of real-time fall detection method based on Martingale Framework will be further increased.

本研究利用传感网络所采集的二值行为数据，对较复杂的日常行为（ADLs，Activities of Daily Living）进行分析和研究，提出二值行为数据的优化方法及传感网络的“顶视摄像机”应用理论，与传统的数字图像/视频处理方法相比，可解决其数据量大、算法复杂及较难实现实时数据在线处理等问题，使在室内环境下应用传感网络小数据量完成较复杂任务成为可能。在前期研究中，我们通过建立移动人物的运动模型，利用准则函数误差最小原则与极值迭代方法，初步提出了人物的动态定位算法，定位精度可达到0.3米（实验环境为15.0米*8.5米）。通过本研究提出的二值行为数据优化方法，人物像素分布与行为强度差异特征更加明显，多行为识别效果可得到显著提升，异常监测鞅框架下的实时监测方法可进一步提高摔倒异常行为的监测准确率。

本研究利用传感网络与可穿戴传感设备所采集的行为数据，对较复杂的日常行为（ADLs，Activities of Daily Living）进行分析和研究。并且较深入的研究了步态特征与脑功能下降引起的退行性病变之间的关系。提出基于MEMS惯性传感的步态特征提取算法，显著提高步态参数的采集精度，有效解决院内步态信息采集的失真问题，实现复杂环境下步态信息的实时自动采集，为临床建模提供客观有效的基础数据。在本研究中，我们已实现步幅等步态参数的高精度采集，误差可控制在2%以内，并且在临床试验中初步展现了步态特征对MCI个体的筛查能力。