Research And Application Of Step Detection Model With Acceleration Sensor

With the rapid development of AI technology,the research and application of intelligent step counting function becomes important.Thanks to the development of MEMS(Micro-Electro-Mechanical System),MEMS-based sensors are becoming smaller and with lower power consumption,and are integrated into various smart mobile terminals,such as smart phones,smart watches,smart handsets,etc.Because of the integration of a series of MEMS-based sensors,it becomes possible for different smart mobile terminals to do something intelligent.One of the most common application is to integrate the MEMS-based acceleration sensor into smart mobile devices.Through the acceleration sensor,mobile device can be us ed to collect the signal data which may be closely related to the user's movement,and by analyzing these data,it can predict the current movement of the user and calculate out the step counts of the user.And the process of calculating the number of step s is so-called step counting.The user's step counts has important application value in the field like motion monitoring,PDR(Pedestrian Dead Reckoning)technology.Therefore,accurately detecting the user's step counts through the signal collected by the acceleration sensor becomes an important research point in the field of step detection algorithm.At present,both domestic and foreign,many scholars have proposed many effective methods in the field of step counting algorithm and motion recognition based on MEMS acceleration sensors.In the motion recognition,most of methods with low self-adaptability due to the fixed placement of the sensors.In the step counting algorithm,most researchers only take methods,such as filtering,correction using gravity,to reduce the influence caused by motion changing on the result of the step counting.Most of these step counting methods rarely combined with motion recognition.In this paper,a self-adaptive step detection algorithm based on motion recognition is proposed,which effectively integrates motion recognition model into step detection algorithm.Different motions lead to different placement of the MEMS sensor,and the corresponding acceleration signal's features will change.With motion recognition model,the step counting system can identify the different motions of human body.Then,the step detection model will effectively adjust the step counting strategies accordingly to achieve the best result of the step detection.Therefore,the proposed method allows different placement of the sensor,and its self-adaptability is not only promoted by data pre-preprocessing.In motion recognition model,in order to improve the accuracy of the motion recognition,some machine learning methods are used to learn and distinguish the features of the acceleration signal,and sub-band energy ratio difference is used to further improve the recognition efficiency.In step detection model,three step detection strategies are designed according to different motion recognition results.Each strategy corresponds to a specific step detection algorithm,which fully embodies the self-adaptability of the proposed method.In this paper,a large number of acceleration data have been collected from different experimenters under the real sc ene.With these true data,we do the experiment on the MATLAB environment.After the experiment,the performance of the motion recognition and step detection algorithm is analyzed.In addition,three other kinds of step detection algorithm are implemented to compare with the proposed algorithm.All results show that our algorithm has better adaptability and the average accuracy rate of step counting is up to 98.9% under various motions.