A Human Action Recognition Method Based on Three-axis Acceleration Sensor

Abstract

A human body action recognition method based on a three-axis acceleration sensor, the invention includes the following steps: 1) obtaining a three-axis acceleration signal output by a mobile device worn by a human body; 2) filtering out noise from the original signal to extract action classification features; 3) using multiple Classifiers are used for classification, and the average of the output results is used as the final output result. The beneficial effect of the present invention is that the accuracy rate of human action recognition is improved, and the accuracy rate of human action classification is higher.

Description

technical field [0001] The invention belongs to the technical field of human body action recognition methods, in particular to the technical field of human body action recognition methods based on triaxial acceleration sensor technology and multi-column bidirectional long-short-term memory artificial neural network technology. Background technique [0002] With the popularization of electronic devices such as mobile phones and tablets, mobile devices with accelerometers included therein are also popular in people's lives. Mobile devices with accelerometers can easily capture the motion of the human body. By capturing the motion of the human body, the sense of user experience is improved, and a good development opportunity is provided for devices equipped with the system. Therefore, the human action recognition system based on mobile devices has received more and more attention from researchers, especially in the field of multimedia big data. [0003] Feature extraction and ...

AI Technical Summary

Problems solved by technology

[0004] For the extraction of time-domain features, the extracted features (human acceleration features) generally only consider the one-dimensional situation, and directly perform feature extraction on the original signal on the time axis. Researchers use the mean value as the feature, and usually use a window function to filter Random peaks and noise, this method has relatively low computational complexity and hardware requirements; there are methods of maximum and minimum values, variance and standard deviation for human action recognition problems, which are quite different for walking and running The recognition of actions has a good effect, but the recognition of actions such as walking in place, walking fast, and going up and down stairs is not very ideal

For the extraction of frequency domain features, most researchers rely on Fast Fourier Transform, Discrete Fourier Transform, and Discrete Cosine Transform to decompose the time domain signal into the frequency domain. To obtain better distinguishing signals in the frequency domain, it is necessary to perform Long sampling time, which will greatly affect real-time performance

[0005] At present, the above two methods of feature extraction are not ideal for human action recognition.

[0006] After feature extraction, it is to classify the proposed features. Existing classifiers such as k-nearest neighbor classifier (k-NN), support vector machine (SVM), multi-layer perceptron (MLP), k-means (k- means), are good classifiers, but these traditional classifiers are not the best choice

Images