Research On Smart Watch System With Fall Detection And Positioning Function

With the development of China’s population aging trend is accelerating, China’s aging process is intensifying. Because of their own reasons, the old people are more likely to cause falls and result in serious injury. There has not been a mature method to establish fall model for the elderly at home and abroad, and the recognition equipment is relatively complex. How to adopt effective pattern recognition methods for fall detection of the elderly has become a research focus in the field of artificial intelligence. The discussion to this question has high research value and application significance.Support vector machine(SVM) has displayed its excellent performance in small sample classification problems and attracted researchers’ attention. The old man’s fall event is a small probability event, so the old man fall recognition using SVM is the very good embodiment in practical application of a small sample of events. Rough one class support vector machine(ROCSVM) is a kind of important algorithm in SVM, the proposed algorithm exhibits excellent properties in terms of the anomaly detection. Because of the ROCSVM training set only the positive class samples, thus fully excavate and use classification features of training samples have an important meaning to improve the classification performance of ROCSVM. Based on the above background and basic theoretical knowledge, in view of the actual problem, namely the recognition problem for human fall recognition, the kernel function in ROCSVM to construct the study. The main work of this paper are as follows:1. Carded and discussed the theory of SVM and the kernel method. Based on the analysis of SVM algorithm, we realize the limitations of the classic SVM on two classification. During the analyzing of ROCSVM, we found the selection of appropriate kernel function would have great impact on the classification performance because the ROCSVM only has a group of training samples.2. In accordance with the characteristics that the ROCSVM training set only has positive samples, we presents a weighted feature-contribution-degree(WFCD) based Gaussian kernel(λ-RBF). The proposed λ-RBF is more suitable for ROCSVM. Experiment on the simulation experiment and the standard database data show that the optimized ROCSVM has a better identification effect.3. According to the proposed λ-RBF ROCSVM is, we designed and implemented a smart system with function of fall alarm and positioning. The recognition algorithm using in the system is proposed in this paper. The recognition results are sent to the server through the communication module and the guardian mobile phone system. If receive fall information, the system will send out alarm information.The system has new feature of convenient wearing and privacy protection. The work we do is a new attempt and breakthrough of fall recognition, which has promising and wide popularization value.