Research Of Human Activity Recognition Based On The Combination Of Distributed Micro Inertial Sensor

The research of human activity recognition is based on visual tools and wearable sensors.However,the cost of visual equipment is high and it is limited by the scope of monitoring in practical applications.With the rapid development of micro-electro mechanical systems,inertial sensors have the advantages of low prices,low power consumption and portability.More and more domestic and foreign researchers have realized the recognition of human activity by inertial sensors.However,current methods have the disadvantages of low recognition accuracy and high complexity of the algorithm.The purpose of the survey is to investigate the method recognizing human activity.Firstly,in this thesis,the principles of human activity recognition based on micro inertial sensors are analyzed.From the feasibility and implementation,the algorithm of human activity recognition based on hierarchical recognition algorithm and the support vector machine algorithm is proposed for the problem of low accuracy and complex algorithm.The experiments have been verified on the platform of micro inertial measurement unit,which is integrated by the laboratory students independently.The inertial measurement units are respectively placed in the testers’ waist and leg.The algorithm accurately identified various activities,including standing,walking,running,upstairs,downstairs,backward walking,forward falls,backward falls and lateral falls.Secondly,the data of accelerometer and gyroscope are respectively processed by smoothing and median filtering before the feature extraction.Because of the large amount of computation in frequency domain and time frequency feature,time domain features of accelerometer and gyroscope are selected as the features of human activity recognition.In order to reduce the feature calculation time,the extracted features should be optimized to reduce the dimension of the features.Time domain features of accelerometer and gyroscope are used to identify the daily behaviors.The abnormal behaviors are detected by threshold,attitude angles and peak value of accelerometer.The support vector machine based on the hierarchical recognition algorithm is used to distinguish the behaviors of walking,upstairs and downstairs,improving the overall accuracy.Finally,after the simulation and optimization of the behavior recognition algorithm on the computer,designing recognition experiments based on single sensor node and multi sensor nodes,the experimental results are verified by the embedded fire soldier positioning system.The results show that the recognition algorithm can recognize multiple human activities.The average recognition accuracy of single sensor is 91.1%,and the average recognition accuracy of multi sensors is 95.6%.For the upstairs and downstairs,the accuracy of multi-sensor nodes is better than 4.5% of single sensor nodes.Therefore,the method of human body behavior recognition based on distributed micro inertial sensor is better than that of single sensor node.The algorithm has a strong practical value.