Research On Key Technologies Of Device-Free Iadoor Localization And Fall Detection

With the development of economy and society,human activities are more and more concentrated in safe and controllable indoor environment.Indoor motion sensing technology can perceive the movement and behavior of people or objects in indoor environment and form a closed loop with the help of a variety of indoor services,which can greatly improve the level of intelligence of indoor environment.Indoor motion perception technology complies with the trend of development,has a broad application prospect in industrial manufacturing,medical care,smart home and other aspects,plays a key role in the new round of industrial reform led by big data,and thus has very important research significance.From the perspective of application,this paper studies indoor location perception and indoor fall perception technology with full consideration of the convenience,reliability,cost and protection of personal privacy of the technology applied.Signal noise analysis,indoor channel model,evolutionary computation algorithms and deep learning algrithms were studied in this dissertation.Taking advantage of low-cost,easy-to-deploy and sensitive Ultra High Frequency(UHF)Radio Frequency Identification(RFID)system and Ultra Wide Band(UWB)radars,a Device-free indoor Localization system with low cost and high accuracy and a Device-free indoor fall sensing system with high sensitivity and high robustness was proposed.Test platforms in real application environment were constructed and used for a comprehensive test and evaluation of the schemes.The main research contents and contributions of this paper are as follows:(1)The indoor transmission path of UHF RFID signal is analyzed and derived,and the indoor signal transmission model of RFID system is established.Firstly,the signal transmission path introduced by the target entering the sensing range is separated by means of difference.Then,after simplification and analysis,the phase difference between the target and the adjacent tag is extracted by using the relation between the adjacent tags and eliminating most unknown parameters through the division of the neighbouring tags' backscattered signal in the complex domain.Finally,by analyzing the cause phase blur,a distance difference constraint is introduced by deploying adj acent tags within half of the carrier wave length range to avoid phase blur.(2)WallSense,a low cost and high precision indoor Localization system based on particle swarm optimization,was proposed.Using the RFID indoor signal transmission model established in this paper,WallSense deploys tag arrays on the wall as sensors and locates the target by collecting the signal strength and phase information of the backscattered signal of the tags.WallSense system first calculated the distance difference from neighbouring tags to the target according to the indoor signal transmission model.Then solve the location of the target using particle swarm optimization method.The particle swarm optimization method assumes that position is known and solve the target location by optimizing the distance measure between observation and the theoretical value of phase difference between neighbouring tags Thus,the method avoids solving the location directly using classic method,which is too complex considering the number of tag pairs.In this paper,the positioning performance of WallSense system was tested by building a test platform in a real environment.The data is collected using real person target and reflection box target The performance of the positioning system is optimized by means of double tag array positioning,objective function weighting and improved subset particle swarm optimization.(3)The positioning problem of Impulse Radio ultra-wide Band(IR-UWB)monostatic radar is studied,and a clutter removal algorithm based on adaptive variance is proposed.Firstly,the original ranging data of UWB radar is preprocessed by Using Hilbert transform.To solve the shortcomings of existing clutter removal algorithms,based on the modeling and analysis of UWB radar signals in complex indoor environment,this paper proposes a method of clutter removal using adaptive variance to determine updating strategies of the clutter map.Based on the decision of the algorithm,strategies such as exponential update or no update is applied.In this paper,the algorithm is tested in a real indoor environment,which proves that the algorithm has good clutter removal effect on stationary and moving targets,and can greatly improve the target detection and ranging accuracy.(4)A method of indoor fall detection using impulse radio ultra-wideband single station radar and cnn-convlstm neural network was proposed.This method makes use of the excellent spatial resolution of ultra-wideband signals,combines the automatic feature extraction capability of convolutional neural network and the time-space feature modeling capability of ConvLSTM network,and realizes an indoor fall detection scheme with high precision.Firstly,the system performs static background removal,wavelet denoising and data enhancement for the uwb radar's original ranging signal,and then uses two two-dimensional convolutional layers to automatically extract the local features of the signal,and then uses one-dimensional ConvLSTM layers to automatically extract the overall time-space features of the signal.Finally,the extracted features are used to train the classifier for classification and recognition.In this paper,a real scene was set up and six types of activities were collected with five volunteers of different heights,weights and genders.A series of tests were carried out on the system,which proved that the system has excellent sensitivity,specificity and accuracy.Tests in a complex lounge environment demonstrated the system's excellent transferability and the ability to detect falls with high sensitivity without retraining in the new environment.