2%。 A method for extracting and reconstructing 3-D maize skeleton was presented based on stereo vision. Firstly, extracts the 2-D skeleton from two images taken from special angles, then uses epipolar constraint algorithm and appropriate matching criterions to match the feature points on the skeleton. The point cloud of the 3-D skeleton is then obtained. After de-noising operation, the final 3-D skeleton model of maize can be achieved by using B-spline curve fitting. Based on the skeleton model, some important agricultural geometric parameters, such as leaf length and the angle between stem and leaf, can be calculated easily. The experimental results show that the method can quickly reconstruct the 3-D skeleton of maize, with a deviation no more than 2%.