Robust human detection through fusion of color and infrared video

Abstract

Nowadays, robust human detection is still a key challenge in the field of Computer Vision. Many people detection systems are based on the use of color cameras. Yet, these cameras have problems when the scene is poorly illuminated or when there are sudden lighting changes. This is why, the use of thermal-infrared cameras seems to be an interesting alternative. Indeed, these cameras show a good performance in cold environments. But they offer many troubles in warm scenarios. Under these adverse conditions, human temperature is similar to the thermal readings of the remaining scene elements. This fact makes it hard to distinguish humans from the environment. This PhD thesis develops and implements a robust multisensor [1] human detection system based on fusing the information provided after segmenting infrared [2] and color videos. The final system has been developed based on the INT3-Horus framework [3] recently created in our n&aIS research team [4].