中文摘要：

跌倒是独居老人最主要的意外风险之一,为快速有效获取跌倒信息,使老年人得到及时救助,提出一种基于Kinect体感传感器的人体跌倒自动检测方法,利用Kinect深度图像技术获取人体深度图像前景图,建立前景图三维包围盒,通过实时计算的三维包围盒的长、宽、高数值以及该数值的变化速度,判断人体跌倒是否发生。利用遮挡融合算法,解决了人体躯干被障碍物部分遮挡时,跌倒事件的检测和判定。在室内居家环境下进行了26种测试场景实验,检测误报率为2.0%~6.0%,漏报率为0~4.0%。该方法可以较为准确地实现人体跌倒自动检测。

英文摘要：

Falls are one of the major risks for the elderly living alone at home.In order to get information of fall quickly and efficiently, an automatic fall detection method using depth image of human body based on Kinect sensor is put forward. Using depth image technology, the foreground depth image of human body is obtained to build the 3D bounding box of the foreground depth image. By computing the length, width and height value of the 3D bounding box and the change speed of these values, the accidental falls can be determined. Meanwhile, when the human body is blocked partly by obstructions, the fall detection and determination are solved by using the fusion algorithm of occluded objects. 26 kinds of test scenarios are arranged in indoor environment, the rate of false positives in the system is 2.0%-6.0%, and the rate of false negatives in the system is 0-4.0%. Expermental results indicate that the proposed method can realize human＇ s fall detection with much accuracy.