Research Of Fall Detection Algorithm Based On Statistical Pattern Recognition

In the global aging society background, the elderly state of health and quality of life in later years needs more attention. Fall in elderly population has a high incidence, and consequences are very serious. Particularly when one falls down in the ground with no help for a long time, it may lead to life-threatening. In addition to taking measures to prevent falling, assistance provided timely after the fall also minimize the bodily injury.This thesis studies how to identify the falling coincidences in everyday life using the features of the falling itself. Based on the investigation of the falling process, an algorithm for recognizing the falling occurring with statistics mode is put forward. The input of the algorithm is the acceleration values of the human body collected by the wireless sensor nodes. In the process of implementation of the algorithm, the acceleration data are transformed into information that reflects the attitude change though modeling. The output of the algorithm is whether the fall occurred.Data sources of fall detection algorithm come from the three axis acceleration sensor placed on the waist. Sensor nodes communicate with sink node by RF, sink node transmit data to PC machine through the serial. All activities are divided into two major categories of fall and non-fall, if the condition of fall meet, the fall occurs. According to the different positions after fall, fall is divided into six types such as fall forward, fall backward, left side fall, right side fall, lying supine and lying prone. In order to distinguish the falling action, this thesis chooses the peak acceleration, angle, energy value as fall characteristics. Three eigenvalues constitute a fall eigenvector. This eigenvector compares with fall training samples to judge whether the action is a fall. If a fall occurs, system triggers the alarm to inform children or community care center, so as to provide rapid rescue action。This thesis has completed the research achievements and innovations are the following:(1) analyze the falling action, establish the human fall model, design fall detection algorithm;(2) selection of sensor and calibration;(3) with VC++as the software platform, develop software to complete data collection, processing, analysis and pattern matching process;(4) design experiments to validate the algorithm and evaluate fall detection algorithm.The experimental results show that the algorithm recognition rate of92.14%is an effective fall detection algorithm. Algorithm’s self-learning function shows better value in the follow-up research and applications.