Abstract
In recent years, an increasing interest in artificial intelligence applications in a widespread spectrum of fields which include, among others, robotics, communications, artistic creations, security and protection technologies, etc., has been observed. Of the latter categories, one field which has largely benefitted is surveillance and security technologies. This fact is combined with an increase in omnipresent automatic surveillance system installations which pave the way to new technologies. Technologies that are being promoted are the ones offering uninterrupted, robust, efficient and reliable operation. In this work, we examine the ability of thermal automatic detection systems to fulfill their role as an essential part of such a mechanism. The primary advantage of thermal detection is the potential to provide a 24-h uninterrupted detection service exploiting its innate robustness against environmental or weather changes and shifts in illumination conditions. For providing a reliable security mechanism, a second requirement is considered sine qua non: the efficiency of the system in order to provide timely alerts for potential threats and incidents. In this work, we evaluate various efficient object detection models operating solely in the thermal/infrared spectrum to examine their role as potential backbone detectors in surveillance systems.