Research And Implementation Of Active Identification Of Human Behaviors Based On Wearable Sensors

Human behavior recognition techniques aim to perceive external manifestations of human behavior and identify their categories based on their perception.It is often used in mobile health care,carry on Sports monitoring,and somatosensory games.The traditional human behavior recognition is based on the behavior perception technology of computer vision,which has the problems of privacy invasion and complex deployment.In the recognition model,all the sensor devices are trained as a whole.When the wearer adds a new sensor,the new sensor must be pre-trained to be capable of recognition which requires a lot of manpower and material resources,and cannot cooperate with the existing sensor devices.In order to solve the above problems,this paper uses wearable sensors to collect behavior data,which has great advantages in the protection of privacy and the patience of the environment.At the same time,combining the idea of migration learning,this paper uses the original sensor nodes to train new sensor nodes,improves the behavior recognition algorithm,improves the flexibility of the cooperation among the sensor devices,and makes the new sensor nodes have the ability of active recognition.The main research and contributions of this paper are as follows:(1)First,combining the idea of migration learning and data fusion,this paper uses the original sensor nodes with the ability of recognition to train new sensor nodes that do not have the ability of recognition,completes the migration of knowledge among the wearable sensor nodes,and trains new sensor nodes to have the ability of active recognition.Moreover,this paper improves the k nearest neighbor classification algorithm and the nearest distance center algorithm by using the five tuple identification information between nodes,thus improving the accuracy of behavior recognition.(2)Second,this paper uses the wearable sensor to collect behavior data,combines the data preprocessing method and the improved behavior recognition algorithm to train the recognition model,and builds the human behavior recognition system by using the SpringMVC framework,completes the real-time monitoring of user behavior activities of wearable sensor devices.