In order to realize the intelligent robot to clearly grasp the 3D information of the surrounding environment, a 3D model reconstruction method based on coordinate transformation is proposed. The binocular stereo vision structure is used to obtain the 3 dimensional geometric information of the object in a single view through image acquisition, camera calibration-correction, stereo matching and 3 dimensional feature extraction. Normalize 3D geometric information in multiple coordinate systems to information in the same coordinate system, and a representative rectangular carton is selected for experimental verification. The results show that the method can realize the reconstruction of 3D model, and it also has certain reference value for the understanding of environmental geometric information and obstacle avoidance navigation.