Human-human Interaction Recognition Algorithm Based On Ultra-Wideband Radar

Ultra-wideband(UWB)radar has the advantages of high range resolution,through-wall detection,ignoring occlusion,and insusceptibility to temperature and environmental influences.It has extensive research in different fields,especially in the field of radar intelligent recognition.With the technical improvement of integrated circuit hardware equipment,the size of ultra-wideband radar is getting smaller and smaller,making it feasible to use indoors.This paper mainly researches six human interaction behavior recognition algorithms based on ultra-wideband radar.UWB radar is used to collect data on human interaction behaviors.Four key features are extracted from the data and input to the support vector machine for identification.Recognition and classification of six human interaction behaviors.Firstly,in view of the problem that there is no radar specific data set at present,this paper refers to a video database UT Interaction data set which is specifically for human interaction behavior,uses p440 model ultra wideband radar to build the radar data set of six interaction behaviors required by this project,and then redundant data processing is done.Then four features are extracted in the two-dimensional time range information domain,and six interactive behaviors are classified and recognized by support vector machine based on improved grid search optimization,and the generalization performance of the recognition model is analyzed.On the basis of studying the two-dimensional information domain,the data in the two-dimensional time distance plane is upgraded to the three-dimensional space of distance Doppler time,and then four kinds of features are extracted.In the case of more one-dimensional Doppler information,the extracted features contain the micro Doppler information of human body,which can be more sensitive to the micro motion information of human body,and make the feature information richer.The training model of classifier more accurate,better recognition performance.In order to verify the recognition performance of the model compared with the two-dimensional recognition model,the comparison results show that the performance of the three-dimensional recognition results is improved by more than 10%,and the recognition performance of the three-dimensional recognition model is analyzed from different perspectives,which shows that the three-dimensional recognition model has good generalization performance.In order to realize the recognition of human interaction behavior of UWB radar,this paper determines the core method of recognition algorithm,starts from the two-dimensional information domain,and extends the algorithm,and researches the data into three-dimensional space.The experimental results show that the recognition model has good generalization performance in both two-dimensional and three-dimensional information domain,but through the two-dimensional recognition model and three-dimensional recognition model compared with the recognition model,the three-dimensional recognition model has better recognition performance,which verifies the experimental conjecture,improves the recognition accuracy,and improves the generalization performance.