Research On Human Activity Recognition Method And Systom Implementation Based On Spatio-temporal Correlation Analysis

Human activity recognition and condition monitoring are important research directions in the field of pervasive computing and artificial intelligence,and play an irreplaceable role in medical monitoring,sports monitoring,and elderly care.With the rapid development of technologies such as the Internet of Things,big data and cloud computing,wearable sensor networks can be closely linked with our daily life,which greatly improves people's living standards.Because wearable devices have features such as portability,ease of operation,and stylish aesthetics,they have become an irreplaceable part of human activity recognition and mobile health.Based on this,the paper systematically studies and analyzes the activities of the elderly on the human activity and health monitoring based on the smart bracelet device.Activity recognition is one of the important components of human activity monitoring.We first proposed the overall program of human activity identification.Connect the smart bracelet with Android smart phones through Bluetooth,and use smart phones to display the human activity results,then upload and save the recognition result to the server,and use the activity identification management system to manage all users and related data.First,the human body wears a smart wristband that integrates a variety of sensors to acquire triaxial acceleration and triaxial angular velocity information generated during human motion.According to the actual application scenario,we set the sampling frequency to 8Hz.In addition,since the data collected by the sensor is accompanied by many random noises,the moving average method is used to smooth the noises.After the data of human activity is collected,the human activity recognition feature vector is constructed using the collected 6-dimensional activity data,and the data preparation work is completed for the subsequent activity identification work.On the basis of constructing the activity feature vector,we used the CNN network,the LSTM network and the CNN-LSTM network for 10 atomic activities such as upstairs,downstairs,walking,running,standing,sitting,falling,washing face,brushing,and washing hand,construct a human activity feature vector classifier and use validation set to evaluate the classification accuracy of the models.After comparing and analyzing the overall recognition rates of the three classifiers,the CNN-LSTM classifier was identified as the recognition model of our human activity feature vector.Based on the above atomic human activity recognition results,we realized the recognition of complex human activities.The complex activities defined in this article include washing,eating,toileting,reading,and sleeping.Among them,the wash activity consists of three atomic activities which include washing hands,brushing teeth,and washing face.First,the data collected of complex activities is used as input to the CNNLSTM classifier to obtain a series of atomic activity sequences.Then,use the dynamic time warping algorithm to match and identify complex activities.Because the data of complex activities is diverse and multi-noise,the recognition effect is not obvious.Therefore,we use the spatio-temporal correlation of complex activities to improve the recognition accuracy.The occurrence of complex activities has a large intrinsic relationship with time and space.We simulate the temporal and spatial data of complex activities and use Bayesian networks to characterize the temporal and spatial correlation of complex activities.