A Study On Fall Detection Based On Bayesian Network

With rapid social development of living standards, the aging population is currently a hard nut facing countries all over the world. Among the elderly, accidental falls is one of the major causes led to injuries and as a result, the invention of protection device designed to facilitate monitoring and convenient to carry will become an important research direction in future.Considering the portability, accuracy and real-time, we study on a design of falling detection system based on wearable devices. The following two parts are the main contents of our work: We implement the long-term human motion prediction by dynamic Bayesian Network; therefore we can save more time for the protection device to start-up. Then we implement the falling detection based on Support Vector Machine, and get a satisfactory result both on precision and recall.When constructing Dynamic Bayesian Network, we learn the model from the motion data by Independence Test. Then the Kalman Filter is applied to make the model adaptive. Al last the whole model is inferred by Loopy Belief Propagation. The Falling detection part is implemented by Support Vector Machine. The Kernel used here is based on Longest Common Subsequence and it is proved more effective for motion classification. Al last, both LDA and grid-search is applied for the optimization for the whole model.we collect experimental data and test our design on the Matlab. The results show that our design could get a relative better recall rate and precession. Most importantly, it does save a lot of time.