Research On Abnormal Gait Recognition Technology For Elderly Care

As our country population aging problem increasingly serious,more and more old people need care and concern.As a result,the elderly care has become a research focus of many experts and scholars.And the gait recognition is an important direction of the elderly guardianship,so it has an important theoretical significance and research value.In this thesis,we will introduce the research status of abnormal gait recognition for elderly guardianship,and systematically analyze the gait recognition technology based on sensor networks,gait recognition algorithm and the algorithm performance evaluation index.On this basis,I put forward an abnormal gait recognition algorithm based on BP neural network and the abnormal gait recognition algorithm based on feature dimension reduction.In view of the traditional gait recognition using a single or a single type of sensors which leads to the identification of the types of gait is few and the recognition precision is not high,this article puts forward an abnormal gait recognition algorithm based on BP neural network.As for the gait data collected,firstly,make a secondary feature extraction,take its 19d characteristics vectors,such as foot's pressure mean value and variance,triaxial acceleration's mean value and variance.And then normalize the 19d feature vectors.At last,build the gait recognition classifier based on BP neural network.It realizes the heterogeneous data's integration,and also can improve the gait recognition rate.Secondary feature extraction of the 19d feature vectors contain redundant information and have higher correlation.Aiming at this problem,firstly,this paper introduces the principal component analysis dimension reduction method.And then,in order to optimum eigenvector,a feature dimension reduction method based on gait characteristics and correlation matrix.Finally,An abnormal gait recognition algorithm based on feature dimension reduction is put forward,and it can improve the speed of the gait recognition classifier.To verify the effectiveness of the algorithm,we build the abnormal gait recognition system based on heterogeneous sensor network,and collected five types of gait of eight volunteers,including limping,tiptoe,flurried,stumbling and normal gait,each gait includes 500 groups of data.The experimental results show that the gait recognition rate based on heterogeneous sensor network is far higher than that of homogeneous sensor networks;The above two dimension reduction methods can both improve the speed of the gait recognition classifier,but the gait recognition rate of the dimension reduction method based on gait characteristics and correlation matrix is higher than that of principal component analysis.