This paper presents a general infrared vision system to be used in robotic applications in natural outdoor environments. In these applications the robustness of the vision system and the automatic settings of the infrared cameras are very important issues. A piecewise linear model of the infrared camera has been identified. This model is used for the design and development of a fuzzy control method by applying visual feedback techniques. The vision system also includes a new fuzzy-multiresolution threshold computation method, which considers knowledge on the application and information on the illumination conditions to select an appropriate threshold for the segmentation of the object of interest. The paper describes the application of the proposed system for surveillance and includes some experiments.