Human motion state recognition based on acceleration sensor

Abstract

The invention provides a human motion state recognition method and a system based on an acceleration sensor. An offline stage and an online stage are divided. In the offline stage, a K-Means clustering method is adopted to build a human motion state recognition model, training and research are carried out based on existing data with a label, and a classification strategy is brought forward; in the online stage, a human motion state recognition real-time system is designed based on an Android mobile phone, and design is carried out from five aspects of data acquisition, data processing, motion recognition, model updating and data displaying; and finally, effectiveness of the clustering algorithm is proved through an experiment. The experimental result shows that building of the human motion recognition model based on the clustering method is feasible, and the model has the advantages of good real-time performance, light weight, easy adjustment and the like.

Description

technical field [0001] The invention belongs to the technical field of data mining, in particular to a method for designing and classifying clustering algorithms in data mining. Background technique [0002] With the rise of mobile Internet technology and the development of wireless sensor technology, sensor data is generated all the time, which contains rich information and has far-reaching research significance. The widespread use of pedometers is one such research outcome. Motion recognition is currently one of the more popular directions. By analyzing the acceleration sensor data, a fixed model of human motion is found. So far, the research has mainly used traditional classification techniques, with little involvement in clustering methods. Since this method adopts a sampling process, all data sets must be known in advance, so it is only suitable for static time series data. And because the time series is fluid and unpredictable, the data is generated over time, and t...

AI Technical Summary

Problems solved by technology

[0005] (1) Motion state recognition based on sound: This method has an obvious defect: the applicable scenarios are very limited and the cost of using this method is relatively high

In the field of smart home research, some people have proposed a sensor network system. By installing various sensors at home, the data is uniformly input to the control platform, and then the specific movements of people at home are analyzed and identified; others fix an acceleration sensor on the waist of the human body. It is very good at identifying 9 kinds of sports such as walking, running, and standing; in some studies, in order to ensure the efficiency of data collection and the reliability of transmission, the acceleration sensor and the storage device are connected through the USB data cable and worn on the body In the end, although it can reach 94% accuracy, it is very inconvenient to wear and cannot be accepted by general users; because people's daily movements are diverse and complex, in order to identify more movement patterns very accurately, some researches Through experiments, personnel placed acceleration sensors on five human body parts including the waist, hips, wrists, thighs, and ankles. The final experimental results can accurately identify more than 20 types of sports. However, due to too many sensors worn, the user experience is extremely poor. Poor, it is difficult to be promoted, but it also shows that multiple sensor data can improve the recognition accuracy

[0010] Since most of the current research methods only consider prior knowledge and ignore the characteristics of dynamic changes in streaming data, resulting in problems such as high accuracy of the constructed classification model under static data and poor actual experience

In addition, current research methods do not take user differences into account, and the trained classification model is heavily personalized.

Images