| Context-awareness is one of the most advancing research topics in the field of human-computer interaction.It is the sensing of human behavior and the discrimination of living environment.Currently many works on human behavior recognition usually rely on sensors and visual devices,such as the use of wearable bracelets to detect human movements through acceleration sensors,etc.However,there exist some disadvantages for these active detections,e.g.,discomfort and privacy leaks.Besides,for the perception of materials in the living environment,previous works generally are based on chemical and physical techniques.Chemical techniques use test reagents to react with materials to identify objects,while common physical ones includes X-rays,ultrasound,etc.,which require special experimental settings.To solve the inherent limitations of existing works such as wearing discomfort,privacy exposure,expensive equipment,and destructive testing,we start research in human identification and material inspection.The main contributions of the dissertation are as follows:1.A human behavior sensing algorithm via wifi signals of commercial wireless devices is proposed.According to the different effects of different actions through the channel state information of wireless devices,we build an action-signal model and send the collected channel state information to behavior-appropriable classifier.The classifier uses the Euclidean distance between data as the behavior template,and continuously converge the action template library while classifying,gradually improves the classification performance.Experimental results show that the average accuracy of the proposed algorithm for the five movements(sitting,standing,squatting,falling and lying down)is above 95%.The model acquires a high accuracy while reducing the false alarm rate to be as low as 2.44%.2.A material inspection method using channel state information of commercial wireless devices is presented.According to the different characteristics of different materials to the electromagnetic wave,a material classification model is established to distinguish different materials.Compared with chemical inspection methods,X-rays,ultrasound,etc.,wireless signals are more universal as passive non-destructive inspection techniques.Based on the guidance of Fresnel zone theory and Rice distribution theory,the presented technique constructs a dynamic experimental setup.We experiment with multiple locations for each material,and integrate the data of all locations into a dynamic heat map.In this way,a more comprehensive dynamic identification fingerprint for static materials is constructed.The cross-validation of CNN and KNN classification algorithms demonstrates that the system has an excellent classification effect.The classification accuracy for ten materials(e.g.lead plate,copper plate,iron plate,fruit juice,etc.)is more than 99%. |
