Reearch And Application Of Behavior Based On Six-axis Sensor

Human activity recognition can be defined as a method of interpreting still images,video,and sensory data to classify a range of human activities.With the development of artificial intelligence and the Internet,human activity recognition plays an increasingly important role in medical assistance and human-computer interaction.In today’s privacy-conscious society,compared with video stream-based human activity recognition,sensor-based human activity recognition has been widely researched and applied due to its ability to record human activity information without intruding into life.The six-axis sensor refers to the sensor that collects data from six coordinate axes in space.Problems such as multi-modal sensor data and noisy and complex input signals have brought huge challenges to behavior recognition.How to design and implement a set of complete,reasonable,and high-accuracy behavior recognition and apply it to actual production and life has important value and meaning.In order to improve the existing problems in the correlation between the sensor contribution weight and the different position input signals in the current behavior recognition algorithm,and proposes an effective behavior recognition algorithm based on six-axis sensors,which makes the model more accurate and performs better,Stable,the model is more interpretable.In addition,an action recognition system is developed and built based on the proposed action recognition algorithm.The main research contents and contributions of this paper are as follows:(1)Aiming at the problems in the behavior recognition algorithm of the six-axis sensor,the distribution of the behavior contribution value of the sensors at different positions is not clear,and the relative relationship between the input signals at different times is not considered,and a combination of relative position coding is proposed.And attention mechanism and behavior recognition algorithm for network optimization(RP-Attn-Opt).This paper first designs a multi-modal sensor weight contribution layer.The input six-axis sensor signal can be calculated by this layer to obtain the contribution level of different sensors to a specific action,and the contribution level feature map can be drawn to enhance the reliability of the model.explanatory.Then the signal after the preliminary feature extraction is input to the global information attention module,which includes the self-attention layer and the relative position encoding layer,and the relative position encoding is introduced into multiple attention heads,so that the model can learn from multiple perspectives To effective location information and feature information.Finally,the algorithm in this paper is evaluated on three public datasets and compared with other classic action recognition algorithms.Experiments have proved that the proposed human activity recognition provides the best performance algorithm proposed in this paper can achieve the excellent performance on three data sets,and the stability and generalization of the model have also been verified.(2)According to the algorithm proposed in this paper,the behavior recognition system is developed and deployed.The system mainly includes the front-end module,which interacts with users in the form of WeChat applets;the back-end management module,whose main function is to process various data and build a bridge to interact with other modules;the behavior recognition algorithm module,based on the proposed Algorithm model training and deployment is the main component of the system;the data storage module is used to store user information and behavior data and recognition results.The system is deployed in the way of Nginx+uwsgi+Flask.By changing the configuration information,the algorithm server can be dynamically updated and upgraded,and new algorithm modules can be added later.The construction and use of the human activity recognition system further verified that the text algorithm model has the meaning and value of practical application.