Human Action Recognition Based on State Detection in Low-resolution Infrared Video

This paper proposes a recognition method of human action based on the states in infrared image frames of thermopile array sensors. Two sensors are used for the top-side deployment. Each frame data collected by the thermopile array sensor is called infrared heat map with a resolution 24×32. It is sequentially processed by the quantification, time-domain filtering and background removal operations. Then those processed data are used for the training of state detection models. The state detection models detect the state of the human target in each of frame, and the state sequence is obtained. Finally, A semantic analysis method functions on the state sequence and converts it into the corresponding action. The experimental results have shown that the proposed method can recognize human actions correctly.