Participatory Fall Detection Method Based On Conditional Random Field

Population aging problem is increasingly serious in our country, the old people’s health and safety monitoring has become one of the mobile computing technology research hot spot.Falling down as one of the important factors that threaten physical and mental health of the elderly, has drawn attention to the researchers from academia and industry.Existing fall detection systems are limited to high computation complexity,needing external deployment, poor extensibility and many other factors, making it can not be used widely. The paper based on the statistics of participatory approach, on both sides of body movement characteristic recognition classification and the use of conditional random field model to judge falling has done in-depth research.First, put forward the classification of action recognition method based on statistical sense.Use sensors built-in smart phones to collect human motion signal, then filter noise to extract peak, covariance characteristics and frequency domain, Use these three characteristics and support vector machine(SVM) method divided human action into daily actions, fall down and suspected action, abnormal gait and static these four categories,which provides the annotation basis for the subsequent fall in the context recognition behavior.Second, the fall detection based on conditional random field model is put forward.According to the classification of human action and the context of the relationship between action, make action and characteristics annotations. By the idea of maximum likelihood estimation and gradient algorithm to train the parameters of conditional random field model, and then in the case of given conditional random field model parameters, use the viterbi algorithm to predict the most likely sequence of actions.Finally, use a variety of smart phones to collect sensor data, experiment on Matlab platform and Visual Studio platform, for the proposed feature extraction, behavior,taxonomy and fall detection method to conduct a comprehensive performance analysis,threshold determination and accuracy analysis.