Design And Implementation Of Personalized Indoor Fitness Behavior Recognition Method Based On WiFi Channel State Information

With the improvement of people’s living standards,the requirements for physical health are increasingly high.More and more people choose to exercise in the home environment to improve their physical fitness.On the other hand,WiFi signal is widely used in indoor action recognition due to its low cost and popularity.Therefore,it is feasible and practical to use WiFi signal to identify the fitness movements in the home environment,which can effectively guide indoor fitness activities.However,there are differences in users’ behavior habits,body types and genders,etc.,and the performance of the training general model used for action recognition among different users is often reduced because these differences are ignored.In addition,a sufficient number of samples and a large amount of time cost are needed to obtain a model with good performance.In order to solve the above problems,a personalized indoor fitness behavior identification model based on WiFi channel state information was proposed.By training personalized recognition models for different users,we can solve the problem of user diversity caused by user differences in behavior,body type and gender.For the sample number problem,the high-frequency sampling frequency is used to collect the data first,then the data is evenly segmented,and the segmented data is fitted to obtain the 95% confidence interval.The virtual data which is several times the original data can be generated by re-taking the value within the interval.In addition,dynamic time normalization is used to measure the similarity between different users,and then the feature data that is most similar to the new user data is migrated from the old user data set,and the different feature data is fused into the data set,so as to minimize the number of samples and time cost required by the new user data collection.Finally,the method of personalized recognition is verified experimentally,and uses three common classification methods: K-nearest neighbor algorithm,support vector machine algorithm,decision tree algorithm,and makes horizontal comparison with the general model.The experimental results show that the classification accuracy of the personalized model is96.6%,92.6% and 93.8% respectively,and the overall classification accuracy is 94.3%,which is greatly improved compared with the 62.3% of the general model.