We present a video stabilization algorithm for a static thermal camera intended for a practical surveillance system, e.g., for border or infrastructure surveillance. In these scenarios, the camera is typically mounted on a pole and is therefore prone to considerable shaking due to high winds. These shakes correspond to erratic misalignment of the video frames and can degrade the performance of detection and tracking algorithms applied on the video frames, as well as annoy the security personnel monitoring the videos on a screen. Thermal images are difficult to process because these images are of low quality, have poor contrast and brightness, contain a smaller number of features, and are noisy. The proposed algorithm, based on a direct method, is simple yet fast and robust, overcoming these limitations and providing considerable frame alignment even in the case of severe camera shake, as demonstrated by experiments with a randomly added shake to the frames as well as with mechanical shake applied to the pole.