Human Action Recognition Based on Inertial Sensors and Complexity Classification

Abstract

In this article, a human action recognition technique based on complexity classification is proposed. Considering the features of human actions such as continuity, individuality, variety randomness, the demands for recognition of different types of actions are different, the problem of action recognition can be classified into simple action recognition and complex action recognition -- the classification criterions are given respectively. Meanwhile, the hardware design of data acquisition device is introduced and the angle variation is chosen to represent the user's body state changes. For simple actions, a real-time recognition algorithm based on template matching performed well on cost control, and a method based on BLSTM-RNN is used for complex motion recognition to improve the accuracy of identification.