A region-margin line segment stereo matching algorithm is proposed to improve the posture recognition precision of a manipulator. The color characteristics of the region and the geometric characteristics of margin line segment included in the image are used to realize an integrate matching，among which the former contains the color and margin line segment，and the latter contains lengths，directions and angles.By making full use of the information of color included in the image ，especially the gradient direction information, this algorithm makes the line segment matching depend on the all information of the image area,not only on its geometric characteristics, such as lengths，directions and angles. Therefore，the criterion of the matching is more accurate than those of traditional algorithms.A posture recognition experiment for a manipulator is undertaken,obtained results validate that the algorithm could recognize the posture of manipulator in higher accuracy in a complicated background, and the relative error reaches 1.7%,which meets the requirements of the attitude recognition of manipulators for the precision.