Weld formation, especially root weld formation is an important criterion to evaluate weld quality and adaptability in Laser-MIG hybrid butt joint welding. A vision sensing system consisting of an external illumination source, a CCD camera, a trigger circuit with the input of MIG current signal and video grabber is established to study the relation between vision signal and root weld formation. The backside image of molten pool is obtained clearly in real time by setting the camera under the workpiece, and the imaging mechanism is analyzed. The image processing algorithm to detect backside width of molten pool, gap width and the offset of molten pool is developed and the calculated result shows a good precision comparing with the actual root width of weld.