Abstract: A reconstruction method for complex object is developed using non-metric images.Firstly,a rotatable platform is designed to capture images and a planar grid board on its top is for camera calibration.With the help of the platform,images are captured and then processed with multi-baseline image matching algorithm to produce corresponding point features.Also benefiting from the rotation platform,the initial value of images' external parameter can be obtained.All these parameters and coordinates of corresponding point feature are seen as initial values for the block bundle adjustment program and then point clouds and accurate image parameters are obtained.In the end,a vision constrained Delaunay triangulation algorithm is used to produce surface model of the object.The method includes artful hardware design,efficient image matching method and rational 3D surface construction approach.Experiment results show that it is feasible and effective.