Over the last decades, several studies have demonstrated the reflectance effectiveness of middle wave infrared (MWIR) for discriminating among different types of vegetation, and estimating the total and leaf biomass of several forest ecosystems. Therefore, a MWIR aerial image capturing system for vegetation observation is urgently required. Furthermore, stitching those MWIR aerial images together is necessary for a panorama of the region of interest (ROI). Most traditional stitching algorithms for aerial images are designed for visible images, and are not suitable for infrared images that are noisy, blurry, or lack detail. In this paper, a novel real-time MWIR aerial image capturing and stitching system for vegetation observation is proposed. The proposed stitching algorithm for aerial infrared images improved from scale invariant feature transform (SIFT) and random M-least square can find a sufficient number of feature points easily and perform rapid calculations. Therefore, the proposed MWIR aerial image capturing and stitching system for vegetation observation can operate in real time.