This paper assesses a monocular localization system for complex scenes. The system is carried by a moving agent in a complex environment (smoke, darkness, indoor-outdoor transitions). We show howusing a short-wave infrared camera (SWIR) with a potential lightingsource is a good compromise that allows to make just a slight adaptationof classical simultaneous localization and mapping (SLAM) techniques.This choice made it possible to obtain relevant features from SWIR images and also to limit tracking failures due to the lack of key points insuch challenging environments. In addition, we propose a tracking failure recovery strategy in order to allow tracking re-initialization with orwithout the use of other sensors. Our localization system is validatedusing real datasets generated from a moving SWIR-camera in indoor environment. Obtained results are promising, and lead us to consider theintegration of our mono-SLAM in a complete localization chain includinga data fusion process from several sensors.