Research On The Theory And Application Of Fall Detection Algorithm For Elderly

Nowadays, the phenomenon that elderly live alone has become more and morecommon with the development of our society. Falling may cause serious injury,especially for the elderly who lives alone, but most of the time, falling do not leaddamage immediately, but the psychological burden after the fall and other environmentfactors, could bring on serious damage when the elderly cannot get timely help. Themain purpose of the fall detection system is to detect the fall of elder in time.The fall detection algorithm is the key part of fall detection system for the elderlyand the research on fall detection algorithm is mainly based on the acceleration data ofhuman body while using data mining methods to detect falls. But a significant point isignored that fall detection is a problem of cost sensitivity, because one miss alarm offalling activity may cause much more serious damage than a non-fall activity detectedas the falling by mistake.To deal with these problems, the main contributions of this paper are as follows:1. We give a conclusion of the existing fall detection algorithms and finda common defect of them that the cost sensitivity issue is not considered,based on the intensive survey of other researchers’ papers.2. The new dispersed features are extracted from the sliding windows of theacceleration series, and it is proved to have the better performance than otherresearcher’s methods on the experiments of fall detection algorithm using featureselection approaches.3. For the first time we apply the cost sensitivity analysis into fall detectionalgorithm and propose two decision rules that using minimal risk bayes methodto deal with the risk function of the single unclassified sample andNeyman-Pearson method to obtain the minimal population risk of all samples.4. Use post validation approach to improve the specificity and accuracy of the falldetection algorithm since the cost sensitivity analysis method pursues highsensitivity while decreasing the specificity and accuracy of the fall detectionalgorithm.