Research On Human Action Recognition Based On Wideband Random Code Radar

Human action recognition has the important research significance and practical value in many fields such as military security,human-computer interaction,and medical care.Compared with contact sensors,video surveillance systems,and thermal infrared imaging sensors,radar technologies transmit electromagnetic waves as probe signals and realize human action accurate recognition by analyzing echo signals and combining machine/deep learning methods,which have the significant advantages of immunity to ambient temperature as well as visibility interference and protection of personal privacy.However,the existing radar technologies have different defects,including the lack of accurate positioning function of single-frequency continuous-wave radar,the high range sidelobes of linear frequency-modulated continuous-wave radar and stepped-frequency continuous-wave radar,as well as the compromise between signal-to-noise ratio and detection accuracy of pulse-Doppler radar.In addition,their echo signals are susceptible to external electromagnetic jamming.Wideband random code signal has broadband and δ-like auto-correlation characteristics.As the radar transmitting signal,it can realize the high range resolution and strong anti-jamming detection.In addition,the signal-to-noise ratio can be improved by increasing the correlation time rather than the signal amplitude.At present,the random code signal radar has not been widely used in human action recognition.In this paper,the wideband random code radar is introduced into human action recognition,and the human action recognition method based on wideband random code radar is researched to realize human action accurate recognition in strong noise environment.The main research work is summarized as follows:(1)The research background and significance of human action recognition were summarized.Based on comparing human action recognition researches by different radar systems,the research status of radar signal processing and action recognition methods were analyzed.(2)An experimental system of wideband random code radar was set up,the generation method and characteristics of wideband random code signal were introduced and analyzed,and the human action recognition method based on wideband random code radar was proposed.In this method,slow time domain ranging accumulation,static clutter removal,and short-time Fourier transform are carried out to obtain the slow time-Doppler frequency diagram of human action.The action feature is futher extracted by the normalization and fast principal component analysis.Finally,the action classification is realized based on the support vector machine(SVM),K-nearest neighbor(KNN),and convolutional neural network(CNN).(3)The human action recognition in free space of wideband random code radar was researched experimentally.Based on the analysis of the slow time-Doppler frequency diagrams and the corresponding first principal component features of eight actions,the average classification accuracies of SVM,KNN and CNN are 96.875%,85.5% and 98.625%,respectively.Moreover,the proposed radar has the potential to recognize multiple angles or multiple human targets.(4)The human action recognition performance of wideband random code radar under the strong noise environment is studied.The results show that the wideband random code radar can reach 27.6-d B anti-jamming tolerance and still can realize the action classification accuracy of 96.39%,which indicate the proposed radar has strong anti-jamming ability and can realize the accurate human action recognition in strong noise environment.