Research On Real-time Gesture Recognition Method Based On Millimeter Wave Radar

The millimeter wave radar gesture recognition technology refers to the use of electromagnetic waves in the millimeter wave frequency band to illuminate the human body.Different gesture movements reflect different electromagnetic wave characteristics,and then recognize gestures.This technology has important research significance and application value in fields such as smart homes,smart vehicles,medical health,and industrial control.Based on machine learning and deep learning technologies,this thesis conducts research on real-time gesture recognition methods in long-range and short-range application scenarios.The main contents are as follows:1.Based on the foundation of radar detection,the millimeter wave radar echo is first modeled,and the principles of range,velocity,and angle measurement are analyzed in detail through theoretical derivation.On this basis,two preprocessing methods for millimeter wave radar are further introduced: static clutter suppression and constant false alarm target detection algorithms.Finally,based on actual needs,gesture actions and corresponding radar parameters are designed for both long-range and short-range application scenarios,and the construction of the dataset is introduced,including the collection scenario,number and composition of the dataset,which provides a data basis for subsequent recognition algorithms.2.For the long-range application scenario,based on traditional machine learning theory,a gesture recognition algorithm based on support vector machine is studied.Firstly,a micro motion suppression algorithm is proposed,which uses the micro doppler effect of the human body to construct a weighting factor and multiply it with the range profile image to suppress the interference of body micro motion on gesture recognition.Then,a foreground segmentation algorithm based on biaxial projection is proposed,which reduces the size of the recognition model while improving recognition accuracy.In order to normalize the same gesture features under different postures,a foreground unification processing algorithm based on directional gradient is proposed,further enhancing the robustness of the recognition algorithm.Finally,the gesture recognition algorithm is deployed and tested on a small embedded system,and typical application scenarios are introduced.3.For the short-range application scenario,based on deep learning methods,a multifeature fusion short-range gesture recognition algorithm based on convolutional neural network and recurrent neural network is studied.First,the principles of typical neural networks are introduced,laying a theoretical foundation for the subsequent network model proposal.Then,a multi-feature fusion gesture recognition network is proposed.The network uses a convolutional neural network to fuse and extract features from the distance-doppler image and the azimuth-pitch angle spectrum image,and uses a recurrent neural network to classify the extracted features which achieves high recognition accuracy under the condition of small model parameters and computational complexity.The effectiveness of the proposed network structure is analyzed through multiple comparative experiments.Finally,the proposed method is deployed and tested,demonstrating the feasibility of the proposed method.