In precision agriculture, yield maps are important for farmers to make plans. Farmers will have a better management of the farm if early yield map can be created. In Florida, citrus is a very important agricultural product. To predict citrus production, fruit detection method has to be developed. Ideally, the earlier the prediction can be done the better management plan can be made. Thus, fruit detection before their mature stage is expected. This study aims to develop a thermal-visible camera system which will register thermal images with visible images, so that information fusion can be done later for detecting immature fruit. The registration method used in this study was based on photogrammetry that could be applied to register multiple cameras as well. The camera system used in the study consisted of two identical visible cameras and a thermal camera which were mounted on a single frame and their positions were fixed. Bundle adjustment was utilized to calibrate cameras’ relative orientations with respect to each other and intrinsic parameters for each camera. Image registrations were conducted in real-time after each set of images were taken by the three cameras. Common points of interest in the two visible images were selected by running the random sample consensus (RANSAC). Coordinates of corresponding points in the thermal image were calculated by utilizing image intersection method. A transformation matrix was then solved based on selected corresponding points in the thermal image and one of the two visible images. Finally, image registration between the thermal image and the visible image was completed by applying the transformation to the thermal image. The stated method is expected to be fast and can be expanded for multiple camera systems