On the basis of traditional solar tracker, a new solar tracker with double modes is built using image processing technology. The theme is focus on sun centroid extraction algorithm for solar tracker. To handle the problem that the solar tracker cannot track accurately when the sun is obscured by clouds or fogs, an image dehazing method based on rapid median filter is proposed. The transmission map is estimated by dark channel prior, and is smoothed using fast median filtering instead of soft- matting. The dehazed image is estimated using the transmission map. In order to avoid segmentation errors caused by manual threshold, an improved fast manimally stable extremal region (MSER) feature regions extraction method is proposed to extract sun area, and the centroid of the sun is computed. The experimental results indicate that the proposed algorithm can restore images captured in foggy and hazy weather, and accurately extract the centroid of the sun. The proposed method has low computational complexity and it meets the real-time processing acquirement of solar tracker.