This paper proposes a binocular reconstruction and monocular 6 Degree of Freedom(Dof) pose estimation for model free robot grasping, mainly consisting of three stages: First, capture images of the estimated object by using a binocular camera to circle around it, and establish a sparse feature point cloud through motion reconstruction methods. Then, during the estimation stage, acquire images through a monocular camera and extract feature points. After that, perform 2D-to-3D feature point matching and Perspective-n-Point(PnP) calculation to obtain the object pose. Finally, achieve a more stable grasping effect by mutually verifying the estimated object poses obtained from multiple identical camera poses and different camera poses. This solution does not require an object model and can achieve a success rate of 85% using only a monocular camera during the grasping stage.