Research On Activity Recognition Method Based On Acceleration Signal Of Human Sensor

First,an algorithm for human activity recognition under autoregressive conditions is proposed.Signal values obtained from sensors and open source databases are output in an autoregressive setup,and activity sequences corresponding to different signal values satisfy a Markov hidden model.The above model is trained using an expectation maximization algorithm and the model parameters and dynamic properties are explored and analyzed for different activities.Finally,an extended version of the duration Viterbi algorithm is used to backtrack the hidden sequences under dynamic planning conditions in order to give the corresponding activity partition intervals using the maximum posterior probability.The experimental results show that the method can achieve good partitioning between different mixed activities with good adaptability in the case of convergence of the model states.The algorithm can also be used for other time series signals,Human activity recognition is an important research direction in human pattern recognition.With the continuous development of Io T devices,human activity recognition of smart hand devices has become one of the important research directions and has a wide range of applications in smart monitoring systems,smart homes,motion detection and other fields.Sensor-based systems,such as smartphones,smart watches,inertial measurement units,etc.,can be placed on different parts of the body as needed,or the activity signal can be measured easily and quickly by the subject carrying around the measurement of daily activities.such as peripheral venous pressure signals and electrocardiogram signals.Second,a deep neural network framework is proposed.A convolutional neural network and a bidirectional long-and short-term memory model are used,which is capable of capturing feature information of sensor time series,and it uses a self-attention mechanism to learn,and select potential relationships at important time points.We demonstrate the effectiveness of the approach on six public benchmark datasets and verify that the performance is significantly improved by combining the self-attention mechanism with deep convolutional networks and recursive layers.Finally,the proposed method achieves a significant improvement in accuracy over the state-of-the-art method between different datasets,demonstrating the superiority of the proposed method in smart sensor systems.Finally,an activity recognition algorithm for extracting key features is proposed.extracted from the acceleration signal collected by the accelerometer,outputs and models the observed sample sequences with a hybrid Hidden Markov Regression model to maximize the retention of feature information between multidimensional signals through key point sequences,the parameters are then optimized with expectation maximization to build the model and segment the states of the data using the Viterbi algorithm.The proposed algorithm preserves the overall characteristics of the data through key point sampling to achieve accurate activity recognition.Therefore,this method can be used to segment and recognize human activity information effectively and achieve accurate detection of human activities.