\section{Conclusion\label{sec:conclusion}}

In this paper, we propose a data fusing and enhancement method to fuse data from multiple low-resolution
thermal sensors and use the SRCNN to enhance resolution, the result shows that
it can improve the accuracy of single frame pose detection by 5\%. 
We also propose a turning over detection method using multiple frames to reduce the noise
and considering the trend of pose change to deal with the residual heat. The result shows
that The accuracy of pose detection is 65\%, and the recall rate and precision of turning over detection
is 50\% and 83\%.