Skeleton-based Human Action Recognition in Low-Resolution Infrared Images

This paper proposes a recognition method of human action based on human skeleton extraction in low resolution infrared images, which are collected by two thermopile infrared array deployed on the side. Each frame of image is processed sequentially through quantification, interpolation and background removal operations, and the original human skeleton data is then extracted using a thinning algorithm. A standard human skeleton model is established and fitted to acquire the standard human skeleton data. Finally, the continuous human action is divided into several separate actions, and an LSTM network is used for human action recognition. The experimental results have demonstrated that the proposed method can achieve an accurate recognition of human actions.

I. Introduction

With the development of society, the high pressure of work and life has caused many individuals to ignore health problems, resulting in an increase in chronic diseases which now become the main threat to human health. Furthermore, due to improvements in living standards and medical technology, people are living longer, and the trend of population aging has been increasing. According to the prediction of the United Nations, China's elderly population will exceed 300 million in 2025 [1]. Based on the needs of social development and the evolving disease spectrum, enlarged health has gradually attracted more and more attention as a global health management concept. Building an efficient and low-cost indoor human action recognition system, perceiving and analyzing human actions in daily activities to predict health risk is an essential part of the transformation of social health management model.