Circular targets used for mobile three-dimensional (3D) vision measurement system have apparent disadvantages such as low and uneven imaging brightness. An approach based on gray difference weighted centroid algorithm is proposed for the targets′ recognition and extraction. Approximate imaging centers and imaging areas of the circular targets are acquired, and then the pixel edge points are detected by scanning the imaging areas in four directions. Sub-pixel edge is located by using gray difference weighted centroid algorithm. And sub-pixel edge points are fitted with ellipse and thus the imaging center positions of the circular targets are acquired. Experimental results indicate that with the proposed algorithm, the elapsed time is about 3 ms, the maximum error of ellipse fitting is less than 0.08 pixel and the average error is less than 0.03 pixel. Compared with other traditional extraction algorithms, it makes great improvement in speed and localization precision, and is more appropriate for the extraction of circular targets used for mobile 3D vision measurement system.