Publication Type:Journal Article
Source:Journal of Defence & Security Technologies, Volume 5, Issue 5, Number 5, p.84-102 (2022)
Keywords:AI, Border surveillance, fragmented occlusion, human detection, machine learning, visual sensors
The FOLDOUT project is concerned with through-foliage detection, which is an unsolved important part of border surveillance. FOLDOUT builds a system that combines various sensors and technologies to tackle this problem. This paper reviews the work done by AIT in FOLDOUT concerning visual sensors (RGB and thermal) for through-foliage object detection. Through-foliage scenarios contain an unprecedented amount of occlusion, specifically fragmented occlusion (e.g., looking through the branches of a tree). It is demonstrated that current state-of-the-art detectors based on deep learning approaches perform inadequately under moderate to heavy fragmented occlusion. Various state-of-the-art and beyond state-of-the-art detection algorithms, based on deep learning as well as on other approaches, dealt within FOLDOUT to detect objects in the case of fragmented occlusion, are presented, discussed, and compared.