Study On Human Posture Recognition Based On Wearable Devices

Human posture includes human behavior of overall body movement and gesture of partial body movement.With the development of Internet of things and other technologies,posture recognition plays an increasingly important role in biomedicine,human-computer interaction,smart home and other important fields.At present,the problem of aging population in China is becoming more and more serious.And the number of sub-healthy people and deaf-mutes in the disabled group is increasing year by year.The design of an effective human posture algorithm and system will help to solve the problems of disadvantaged groups,especially the problems of home-based care for the elderly and the communication barriers of deaf-mutes.There are privacy and environmental interference problems in traditional behavior recognition and sign language(SL)recognition based on visual image analysis.On the one hand,the existing wearable device based behavior recognition system has problems of few recognition categories and low accuracy.On the other hand,the existing wearable device based SL recognition system has problems of high user relevance,low accuracy and few categories.In view of the above shortcomings,we conduct research on human posture recognition based on wearable devices,mainly focusing on human behavior recognition and sign language recognition in posture.On this basis,we design a behavior recognition system and a sign language recognition system respectively.The detailed study is as follows.(1)Study on human behavior and sign language characteristics.This paper discusses the research object of human behavior and gesture in human body postures,analyzes and summarizes the characteristics of human behavior and gesture.Human behavior can be divided into abnormal behavior and daily behavior.Chinese sign language(CSL)in gestures can be divided into alphabet,number and word SL.(2)Design of behavior recognition algorithm.In view of the existing problems of fewer categories and low accuracy of behavior recognition based on wearable devices,we extract the characteristics of human behavior including human tilt Angle,percentage of height difference,plantar pressure and human acceleration.We design threshold based human behavior recognition(T-HBR)algorithm and conducted offline experimental test through the collected experimental data to verify the performance of the algorithm.We also design plantar-pressure based stride frequency calculation(PP-SFC)algorithm to detect footstep frequency during walking or running.(3)Design of sign language recognition algorithm.In view of the existing problems of high user relevance,low accuracy and few categories of users in SL recognition based on wearable devices,we make a preliminary study on the pretreatment of SL data.Pretreatment includes normalization,digital filtering,signal feature extraction and physical feature extraction.We also design curvature and open fuzzy classifiers(CFCL and OFCL)to extract the physical characteristics of the curvature and the openness of fingers.Furthermore,we construct fuzzy-classifier based multiple classifiers of single CSL(FC-SCMC)using CSL data features and support vector machine method to identify number and alphabet CSL.What's more,we design string-matching based compound CSL word recognition(SM-CCWR)algorithm using the characteristics of CSL to recognize compound word CSL.Finally,we conducted offline experiments with the collected experimental data to verify the effectiveness of CFCL and OFCL fuzzy classifier,FC-SCMC model and SM-CCWR algorithm.(4)Design and physical experiment of wearable device system.We design the behavior recognition system and SL recognition system based on wearable devices,hardware circuit,data acquisition and communication sequence.We also test the systems in physical experiments and compare them with other papers.We design a wearable device based behavior recognition system to verify the behavior recognition algorithm,which can recognize 11 types of human behaviors.And overall accuracy(OA)is 98.0%.What's more,we design a wearable device based SL recognition system to verify the SL recognition algorithm.The OA of 10 number and 30 alphabet CSL is 99.80%and 99.0%,and OA of 10 compound word CSL is 97.87%.