Behavior Analysis Technology Based On Radio Signals

With the rapid development of computation technologies,it is very important to analyze and understand human behaviors.Recently,radio based behavior analysis technology has attracted increasing attentions.The rationale behind radio behavior analysis is that it can infer the activities of human from the fluctuations of the radio signal.It can work remotely and device-free without additional devices.Currently,radio based behavior analysis technology has been deployed in many applications such as daily activity recognition,identification and bio-signal detection.However,due to the dynamics of signal propagation and the complexity of the surrounding environment,there are still many challenges.(1)Radio signal is non-stationary and non-linear.It is hard to extract fine-grained features from the signal utilizing the traditional methods such as statistics,short time fourier transform and wavelet transform.(2)Radio signal is heavily affected by the environment,classifiers that are trained using data collected from one scene can hardly work well in a new scenario.(3)Most researches focus on the detect and recognition of the basic actions.However,as for behavior analysis,we need to get an deeper understanding of the actions.To tackle these challenges,this paper explores innovative algorithms and methods on improving the recognition accuracy,conducting cross-domain sensing and deepening the analysis level.The fact that radio signal is non-stationary and non-linear makes it different for traditional algorithms to extract fine-grained features effectively,thus decreasing the recognition accuracy.To address this issue,an empirical mode decomposition based algorithm is proposed to obtain intrinsic features from original signal.This algorithm can effectively extract gesture feature from the CSI streams.Also,we build a gesture recognition prototype Wi Gesture based on commercial WiFi devices.Extensive experiments show that WiGesture can recognize gestures accurately and outperforms the wavelet based algorithms by 16%.The intuition of radio based behavior analysis is to infer specific actions from the distorted radio signals.However,radio signal is not only affected by the actions but also the surroundings.Any changes in the environment may cause severe consequences on the signal.To investigate how the environment affects the radio signal and how the changes of environment affect the sensing results,we constructed diverse sensing scenarios.A large number of experiments are carried out to get an comprehensive understanding of the effects that environment factors such as wall,obstacles,locations and testing person may have on the radio based analysis.To resist the decrease of accuracy during scenario variations,we proposed Wi Hand,a location-independent gesture recognition system based on commercial WiFi devices.With adaptive subcarrier selection algorithm,WiHand can always choose the most affected subcarrier which contains useful information on the gestures.With low rank and sparse decomposition algorithm,WiHand can separate the gesture signal from the complex background information,thus making it possible to extract effective features which is closely related to the gestures.Plentiful of experiments suggest that WiHand can achieve high accuracy under various scenarios without retraining.Most existing research on action recognition has focused on discriminating between different actions,however,the quality of executing an action has received little attention thus far.We study the quality assessment of driving behaviors and present WiQ,a system to assess the quality of actions based on radio signals.This system includes three key components,a deep neural network based learning engine to extract the quality information from the changes of signal strength,a gradient based method to detect the signal boundary for an individual action,and an activity based fusion policy to improve the recognition performance in a noisy environment.By using the quality information,WiQ can differentiate a triple body status with an accuracy of 97%,while for identification among 15 drivers,the average accuracy is 88%.Our results show that,via dedicated analysis of radio signals,a fine-grained action characterization can be achieved,which can facilitate a large variety of applications,such as smart driving assistants.