论文部分内容阅读
In recent years,the Wi Fi sensor constrains the Human Action Recognition scheme to an immovable location when providing training samples,drastically reducing practical application.Furthermore,a location-independent method to recognize human action poses an even more significant challenge due to the ubiquitous experience of participants in environments with wireless signals.Therefore,it is challenging to develop a method to recognize human action irrespective of location.Wi Fi sensing enables many innovative applications such as healthcare,security,etc.Channel State Information（CSI）from packets is widely used in machine learning（ML）to classify events.These algorithms,however,require much computational power to extract and collect additional CSI features.The method of human action recognition is studied in this thesis,which is based on WIFI channel state information（CSI）.Firstly,a location-dependent human action recognition（LDHAR）method is presented.Then,the CSI data is analyzed,and the human action can be categorized according to CSI remotely.However,the shortcoming of this kind of method is that it requires extensive data collected from a fixed environment.Hence,discrete wavelet transform（DWT）is adopted to decompose the CSI data.Given DWT,the features of human action can be analyzed through shallow algorithms.The author then identified the location-dependent shortcomings of CSI sensing environments which required extensive data collection from a fixed location.In our second work,which is based on the CSI data from two locations,a 1D CNN-LSTM model and antenna sectors comparison is developed for an innovative concept called Dynamic Sensing LocationIndependent Human Action Recognition（DS-LIHAR）.Our proposed work extracts the CSI of Wi Fi to predict actions samples more accurately,allowing for a more time-effective data collection process when compared to a location-dependent environment.Thus,the method reduces the impact of human action recognition on location independence.The experimental results reveal that,when adopting our proposed strategy,DT can obtain over 94.9% accuracy for location-independent human action recognition and over 90% average accuracy for CNNs.Furthermore,the location-independent model developed can be used for practical applications in any situation,and data from each antenna is processed independently,making it applicable in any location or antenna setup.Furthermore,sample lengths of various inputs have been tested to overcome the measurement sensors’ sampling rate limitations.It has been proven that wireless signals can be utilized to detect and track the movement of human beings.The suggested Wi Fi CSI-based location independent approach for identifying human actions,as a result,could serve as an excellent platform for locating human actions in today’s technologically advanced world.