Research On The Method Of Human Motion Recognition By Radar Based On Deep Learning

In recent years,scientific and technological research and precision manufacturing processes have continued to progress,the accuracy of radar measurement has become higher and higher,and the cost of radar has gradually decreased.Radar applications fields have gradually spread from military scenarios to civilian scenarios.Radar has incomparable characteristics of other sensors in the field of motion recognition applications.The basis for human motion recognition based on radar is that changes in limbs will cause changes in radar echoes.The higher the radar resolution,the more obvious the echo changes,so High-resolution radar can distinguish the subtle differences of different human movements to achieve better recognition performance.This thesis is based on carrier-free ultra-wideband radar and Frequency Modulation Continuous Wave radar commonly used in industry in recent years to collect human motion echo signals,extract key features,The recognition algorithm uses more intelligent machine learning and deep learning algorithms.The main research contents are as follows:1.Establish hardware platforms based on carrier-free ultra-wideband radar and FM continuous wave radar,respectively,and design practical actions for home health monitoring and human-computer interaction,and use radar hardware platform to collect motion echo signals.To process and generate action datasets respectively for verification of subsequent action recognition models and algorithms.These types of actions are more typical actions in the corresponding scene.2.Introduce the radar echo model of human motion and the theoretical basis of radar motion recognition.Based on the Distance-Doppler figure,analyze the motion echo signals from two angles,and introduce the micro-Doppler characteristics and Time-Distance feature,Based on the theoretical basis of these two motion echo signal characteristics,Summarize a universal human motion recognition model using radar.3.A human body motion recognition method based on Time-Distance feature and deep learning algorithms is proposed.Utilizing the high range resolution of the carrier-free ultra-wideband radar and the dynamic characteristics of human motion,the two-dimensional features of the distance and time of the human target were extracted to make up for the shortcomings of the single range feature.Finally,an optimized convolutional neural network is used for feature map recognition.Based on this theory,the SIR-20 high-speed ground penetrating radar platform was used to collect data and recognize 8 different human motions.Finally,average recognition accuracy of 99.2% was achieved on a data set consisting of 8 human actions,Explain that the proposed method can accurately identify actions.4.Propose a human motion recognition method based on micro-Doppler features and dictionary learning.Firstly,the Frequency Modulation Continuous Wave radar is used to collect the radar echo signals of human motion,and Time-frequency analysis of the echo signal to obtain the time-frequency map.Then,two feature extraction methods are used to describe the time-frequency map,and the two dimensionality-reduced data are fused.The LC-KSVD dictionary learning algorithm learns a multi-feature dictionary and a linear classifier at the same time,and finally recognizes actions based on the sparse coefficient and the linear classifier.Based on this,a 77 GHz millimeter-wave radar motion recognition experimental system is designed.The results show that 97.7% recognition accuracy is achieved on the 10 human motion datasets.It can be seen that the proposed method achieves accurate recognition of human motion.