Easy rectification for infrared images

Most applications using infrared thermography only take advantage of one feature in the images: the intensity of the objects in the infrared images, which is mainly a function of its temperature. Many different applications use this feature as an indicator of health, early signs of malfunction or signs of hidden conditions. However, infrared images also contain relevant geometric information that can be used to measure objects or to locate areas of thermal contrast in the scene. The problem is that the extraction of geometric information requires a complex camera calibration procedure that depends upon calibration plates which are difficult to build. In this work, an easy rectification procedure for infrared images is proposed without using calibration plates. The proposed method uses a camera projection model not considering distortions, which greatly simplifies the estimation of the projection parameters while producing very good accuracy. The method estimates the projection parameters iteratively based on features from objects in the image and the knowledge about its geometric properties. The result is a method that provides reliable geometric information about the objects in the scene with a single image. A series of experiments are performed to validate the proposed method. Results show excellent performance, with sub-pixel accuracy.