Aiming at the filed of robot grasping using binocular vision, a new method for dealing with the depth information is proposed in the multi-target scenarios. By setting a threshold in pixels and traversing the depth information image similar to clustering method can obtain the coordinates of the object and pass the coordinates to the robot arm to grasp the object accurately. In this paper, the principles of binocular vision, camera calibration, stereo recertification and stereo matching are described. Their difference between the original depth information and the depth information after optimization can be acquired by utilizing the optimization algorithm. The approach for grasping object for robot in multi-target scenarios is proved to be effective in the experiments. Research directions for future improvements are also presented.