Inertial sensor based non-feature human motion recognizing method

Abstract

The invention provides an inertial sensor based non-feature human motion recognizing method. The method comprises the following steps: (1) fixing an inertial measuring unit at the back of a waist of ahuman body; and collecting inertial data of six motion modes (including static standing, walking, running, jumping, going upstairs and going downstairs); (2) preprocessing the inertial data; (3) performing single-circle splitting on data through a zero-crossing detection method; (4) generating a JPEG image through the split inertial data; (5) treating the generated image as input, and re-trainingan Inception V3 model supplied by Google based on a transfer learning method; and (6) modifying the last complete connecting layer of the Inception V3 model, and forming the non-feature human motionclassifier. According to the method, the complex process of manually extracting features is saved; the inertial data are used for generating the image; the features are automatically extracted based on the CNN network, and the classifier is formed, so that the time and the labor cost are saved; and the problem of low recognizing rate caused by inaccurate classifying feature selecting can be avoided.

Description

technical field [0001] The invention belongs to the field of human motion pattern recognition, in particular to an inertial measurement unit worn on the human body to collect accelerometer and gyroscope signals, and propose a featureless human body motion pattern recognition method based on the obtained time domain signals. Background technique [0002] Inertial sensors are one of the common data sources for human activity monitoring and motion classification. Numerical methods are usually used to analyze and process the time series data generated by sensors. Machine learning techniques such as random forests and support vector machines are often used for motion classification, but they need to manually extract various potential features from a large amount of raw data, and signal preprocessing and feature extraction steps require professional and complex digital signal processing technology. [0003] Inertial sensors are ubiquitous in everyday life, such as mobile phones ...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to improve the existing human motion recognition method, which requires manual extraction and screening of features, resulting in a large amount of algorithm calculation and low recognition rate. Using the inertial data collected by the waist sensor, combined with computer vision and deep learning algorithms, a new method is proposed. A Featureless Human Movement Recognition Method Based on Inertial Sensor

Images