Research And Design Of Health Monitoring Technology Based On Multi-sensor

As human society enters the age of aging,falling over has gradually become one of the major detriments to the aged both mentally and physically.The technology of fall detection and behavior recognition has gradually become the research hotspot by scholars.However,the detection accuracy of a single sensor is not ideal since it is affected by the lighting,blocking and other factors.It is an important development direction of human fall detection and behavior recognition technology using multiple sensors.This study revels around the research of fall detection and behavior recognition algorithms in multiple lighting environments,which centers on the algorithm of falling detection and recognition through the single sensor and multi-sensor for human motion.For this purpose,this paper carried out these works using the method of theoretical analysis,algorithm research and measured verification.The main contents are as follows:1.A deep learning network for adaptive feature extraction and classification is designed in this research.Based on the theory of deep learning,the improved convolutional neural network and the long short-term memory network are used to extract the data's features collected by sensors and to learn the space-time and timing characteristics,training the behavior classifier.It solves the problem of incomplete feature extraction in traditional methods and improves the accuracy of single sensor recognition algorithm.2.A recognition algorithm of refined motion based on millimeter-wave radar spectral data fusion is presented in this research.The data of time-frequency and time-distance which includes the information of human motion is obtained,using the method of radar data pre-processing and spectral data analysis.The characteristics which have the information of human motion from multi-dimension are extracted from multi-modal fusion data,which solves the problem of low recognition accuracy for confused action under single mode data.Furthermore,it also improves the accuracy of fall detection and daily behavior recognition.3.An algorithm of fall detection and daily behavior recognition based on the data level fusion of multi-sensor is proposed in this research.A conversion coordinate system is established for multi-sensor including millimeter-wave radar and optical camera to obtain the data after space-time alignment.When the fusion data has completed the analysis of time series feature,a deep learning network is designed to extract the spacetime and timing characteristic information in multimodal data,solving the contradiction between data variability and fusion effectiveness in multimodal data and the problem that lighting conditions affect camera recognition accuracy.Meanwhile,the accuracy in a low light or dark environment has improved because of the proposed fusion algorithm.The measured date is used to verify the effectiveness of these algorithms proposed in this study.The result of confounding actions algorithm's accuracy has shown a 4%increase based on the multi-modal data from millimeter-wave radar sensor.The accuracy of the detection and recognition using camera in normal light environment is above 98%,and the recognition accuracy in dark environment is increased by above 7%.Finally,the accuracy of the data-level fusion algorithm was more than 95% in multiple lighting environments.