Research On Human Behavior Perception And Classification Algorithm Of Low-Cost Radar Sensor

Intelligent perception and classification of human behavior is a research hotspot in the era of intelligence.Although traditional visual sensors or infrared sensors can achieve the above functions,they are expensive and easily affected by factors such as light and weather,and there are privacy and security issues.Therefore,we propose a human behavior perception and classification algorithm based on low-cost radar sensors.Through low-cost radar sensors,coupled with appropriate radar signal processing algorithms and target classification algorithms,we can detect human behavior with high accuracy.Perception and classification,and lower cost,more suitable for practical applications.The main content and innovative research results of this paper are as follows:(1)A Kinect device-based human daily behavior modeling and radar echo signal simulation method is proposed.Firstly,the radar echo signal model of the human body is constructed,and then the optical sensor provided by the Kinect device is used to collect the three-dimensional motion data of the daily behavior of the human body,and then the conversion of the three-dimensional motion model to the radar echo signal is realized.The results show that the radar echo signal generated by the above method is completely suitable for the design and debugging of human behavior intelligent recognition algorithm,which simplifies the data acquisition process and improves the overall efficiency of algorithm development.(2)A low-cost continuous wave radar-based fall detection algorithm for human indoors is proposed.Firstly,through the time-domain energy threshold detection,the suspected fall action is screened,and the signal with a large difference from the fall action is filtered out,which avoids the waste of computing resources.Then the STFT algorithm is used to extract the falling action that may contain the Doppler frequency and time relationship,and then the features are classified and recognized by the combination of the PCA algorithm and the SVM algorithm.To verify the performance of the proposed algorithm,a data set containing three different subjects was created,and the data set was augmented with simulation data.The recognition accuracy was calculated using 5-fold cross-validation.The results show that the method has an average accuracy rate of more than 93%for the recognition of indoor human fall movements,and requires less computing resources,lower overall cost,and is more suitable for practical applications.(3)A human gesture recognition algorithm based on low-cost FSK radar is proposed.Taking full advantage of the advantages of FSK radar,the time distance information and time speed information of different gestures are extracted.And through the mapping on the time scale,the distance and speed are combined into the third-dimensional distance and speed information,and the three-dimensional information is used as the three channels of the RGB image to form the final RGB image carrying the three-dimensional information,which is sent to a lightweight convolutional neural network for further processing.Final gesture classification.To verify the performance of the proposed algorithm,a dataset consisting of 5 common gestures was created,and the recognition accuracy was calculated using 5-fold cross-validation.The results show that the recognition accuracy of 5 common gestures is above 98%.