Action identification method based on wearable device

Abstract

The invention discloses an action identification method based on a wearable device. Data is collected by using a single inertial sensor node, and compared with the scheme of multiple nodes, the actionidentification method has the advantages of being more convenient and comfortable to wear. Complete motion data are automatically sliced via a differential threshold detection method, thereby effectively reducing the interference of non-action data. Based on the analysis of five kinds of action features, a standard mean value, a standard deviation, a kurtosis, a skewness and a minimum value are extracted to serve as identification and classification feature values, and dimension reduction processing is performed on the features to reduce the redundancy of feature information. For the linear indivisible problem, a simpler kernel function is designed to reduce the computational complexity. A multi-classification process is implemented by multi-level SVM. Compared with the traditional 1V1 strategy, the design has the advantages of reducing the number of sub-SVMs, meanwhile adjusting the classification priority according to the size of the gravity center distance to improve the classification accuracy, and an effective scheme is provided for a wearable action identification design based on inertial data features.

Description

technical field [0001] The invention belongs to an action recognition method based on a wearable device, and belongs to the technical field of pattern recognition. Background technique [0002] With the development of microelectronics technology and artificial intelligence, intelligent electronic devices are becoming more and more popular in daily life, and the recognition of human body movements by devices has become a new type of human-computer interaction. In the existing action recognition system solutions, the action recognition method based on video analysis uses the motion feature information recorded in the video to analyze the action behavior. Among such applications, the Microsoft Kinect somatosensory game control platform is well-known. This type of method has a complex structure and data The amount of calculation is large, and the application scenarios are limited, so it is not suitable for portable wearable devices. In contrast, motion recognition based on iner...

AI Technical Summary

Problems solved by technology

[0003] After searching the prior art literature, it was found that S.Karungaru et al published a paper titled "Human action recognition using wearable sensors and neural networks," on the 2015 10th Asian ControlConference (ASCC), which used multiple motion data acquisition nodes Collect motion data and use neural network algorithms to recognize human body movements, but devices with more than five nodes are too complex and are greatly limited in daily applications

[0004] Another search found that F.T.Liu et al. published a paper titled "Gesture recognition with wearable 9-axissensors," at the 2017 IEEE International Conference on Communications (ICC). The vector machine (SVM) method classifies and recognizes actions, but this method only makes some optimizations for feature extraction, and uses a simple window segmentation method for action data extraction, which cannot rule out the interference of data in non-stationary states, and in SVM multi-classification The traditional 1V1 strategy is used on the Internet, and there are deficiencies in classification speed and computational complexity

Images