Device-Free Activity Recognition Based On The Quality Information Of Received Signal

With the development of modern information technology,the Internet of Things(IOT),which fuses various technologies such as sensors,information processing and communication networks,raise the sensing ability to a new height.A variety of intelligent devices,such as intelligent refrigerators,intelligent loudspeaker boxes and so on,are gradually walking into thousands of families,they provide the new,intelligent high quality services for the people.As the major developing trend of the future intelligent life,human activity recognition based on all kinds of sensor technology has been one of the research focuses in the signal processing and the relevant fields.The early human activity recognition mainly relied on the wearable devices to carry out the real-time capture and sense all kinds of human postures.However,it is difficult for people to take along the wearable devices in real time in the bathroom and other special environment.Therefore,Device-Free Activity Recognition(DFAR)has caused the wide concern from the research scholars in the relevant fields as well as the application manufacturers.Machine vision,infrared,radar,radio frequency and other relevant technologies all can realize DFAR.Radio frequency technology has the low power consumption,high accuracy and other characteristics,therefore,this paper mainly researches on Activity Recognition based on Radio Tomographic Imaging(RTI).RTI is a kind of emerging imaging technology according to the mutual relationship between the wireless signal and monitoring scenario,it mainly acquires the decrease change information of wireless signal in the observation area through the wireless sensor network,and obtains the images of observed scenario by the inversion process.Since RTI is a classical pathological inverse problem,it is very easy to be impacted by the observation scenario,and the error is higher.This paper constructs RTI images and rebuilds the model based on the famous Compressed Sensing(CS)to explore RTI Activity Recognition technology through researching the Cosparse Analysis Compressed Sensing method.First of all,this paper makes a comparison and analysis for the basic performance of RTI restructing algorithm of Tikhonov regularization,Total Variation(TV)regularization and Singular Value Decomposition(SVD),to confirm the Frequency Domain Correlation(FDC)between the restructing images and reference images as well as the relationship between the index and RTI location error.Secondly,we introduces Cosparse Analysis Model(CAM)in the Compressed Sensing theory to model the system of RTI,and use Greedy Analysis Pursuit(GAP)Algorithm to Solve the inverse problem.At last,it constructs DFAR system based on CC2530wireless sensor node of TI Company,and carries out the verification of the algorithm.The experimental result shows that FDC index of GAP restructing algorithm enhances0.0546than the traditional Tikhonov methods,the system average location error reduces0.02m,the recognition accuracy of squating and laying action respectively enhance 0.12 and 0.02.Thereby it has confirmed that the scheme proposed by this paper can effectively enhance the quality of restructing signal,reduce the location error and improve the accuracy of system recognition.