A new pixel matching method is proposed based on two-dimensional empirical mode decomposition (2D EMD). It combines with five steps of unequal step algorithm and is applied to the on-line three-dimensional shape measurement. A sinusoidal fringe is projected on the measured object moving with a constant speed. The object′s movement can produce the corresponding phase shifting deformed patterns. And five arbitrary deformed patterns are captured by CCD within a duty cycle of the grating pattern. The deformed patterns are decomposed with 2D EMD method. The mode patterns which is consistent with the object′ surface shape are obtained. And then Otsu algorithm is used to do the binarization. The higher reliability of the area is segmented to be regarded as a template to realize the pixel matching. And the equivalent shifted phase deformed patterns in which the points are one-to-one corresponding are calculated. Then, the five unequal steps phase shifting algorithm is used to obtain the wrapped phase, which forms unwrapped phase by the phase unwrapping algorithm. And the surface information of the measured object is retrieved via the phase-height formula. The computer simulations and experiments prove the feasibility of this method.