To address the ambiguity of information and uncertainty in the image-guided application, a Multi-source information fusion technology is proposed for infrared/visible dual-mode compound imaging guidance. With augmented variance ratio of two classes, an on-line adaptive feature selection scheme is applied to selecting the distinguished attributions in the infrared and visible images. The results of object tracking are yielded in infrared/visible images respectively based on a particle filter algorithm; by introducing the quality metrics factor of tracking in single-mode image sequences, a weighted fusion strategy is employed to achieve the robust tracking in image sequences, the experimental simulation results has verified the effectiveness of the scheme.