Research On Human Motion Recognition Based On Wi-Fi Signal And Inertial Sensing Data

With the rapid development of computer software and hardware technology,people are no longer satisfied to interact with computer through command lines or graphical interface.Instead,they pursue more natural and friendly interaction with the computer directly through human motion at any time and place.Therefore,human motion recognition has attracted more and more attention from researchers and became one of the research hotspots in the field of human-computer interaction and pervasive computing.Human motion recognition research uses data processing,pattern recognition,machine learning and other related technologies to allow computers to acquire data and understand human motion information.Then we can achieve higher levels of human-computer interaction.It has extensive research and application in sports analysis,health care,special education and other fields.Human motion recognition mainly collects the human motion data through the related facilities,and then pretreats the data such as de-noising.Next,we choose the appropriate feature extraction method to extract feature values that can characterize the motion.Finally,according to the feature,we can classify and recognize human motion.In recent years,the methods of human motion recognition mainly based on vision,inertial sensors and wireless signals.The vision-based motion recognition method has already had many mature research results,and it has been able to achieve a good recognition effect on the basis of the existing image database.However,it is greatly affected by the environment,and it is difficult to achieve accurate recognition in the case of insufficient light or long sight distance.The recognition method based on inertial sensors avoids the interference of environmental noise,and does not afraid of blocking.It also reduces data processing.However,it needs to wear the device on the body,which not only increases the cost but also brings inconvenience to the activities.The wireless signal-based identification method can collect action data at anytime and anywhere,and do not have to carry any equipment.However,it is easy to interfere with the noise of the environment,and the identification system with a high recognition rate needs the assistance of a dedicated hardware device,which is not conducive to large-scale deployment.In this paper,the multimodal data fusion of Wi-Fi signal and inertial sensing data is proposed to identify the human motion with fast moving speed.Collecting Wi-Fi signals and wearing an additional inertial sensor on the arm to obtain motion sensing data,we can recognize 7 basic table tennis movements respectively with these two kinds of data.After correlating and combining two kinds of data,we used the fusion information to build the model of the research object,and then get more accurate identification results.The main contents of this paper are as follows:1?Processing primary data: We establish a wireless data acquisition system through an existing Wi-Fi device and a commercial wireless network card to obtain channel state information(CSI)data.We smooth de-noising the data reduced by principal component analysis to reduce the interference of extraneous noise,and use wavelet packet decomposition technology to extract the energy value of wireless data samples as the eigenvalue sequence.We use Fast Fourier Transformation(FFT)to process inertial sensing data,extract abnormal data through spectrum analysis and setting thresholds,and transform the data to obtain the feature value sequence.2?Extracting feature vectors: We use the K-means clustering algorithm to vector quantize the feature value sequence to obtain the feature vector.While retaining sufficient feature information,we reduce the dimension of the feature sequence.3?Modeling and recognition: We set up a Hidden Markov Model(HMM)to classify and identify the two kinds of data.Then,we use the Back Propagation(BP)algorithm to fuse the two kinds of data and identify the motion.Experiments show that the human motion recognition model based on Wi-Fi signal and inertial sensor data fusion proposed in this paper can effectively identify 7 predefined table tennis basic movements under non-massive data samples.The average recognition result is higher than that of using Wi-Fi data or inertial sensing data individually.The average recognition rate of human motion can reach 97.43% after the fusion of data.We verify that the fusion method has certain stability and feasibility by randomly selecting training samples and changing the number of training samples.It has a certain research value in the application to the reality,and provides a new idea for the relevant issues of motion recognition.