Compared with model-based manipulator calibration method,model-free calibration method has the advantages of simple algorithm and small linearization error,but this method requires the simultaneous acquisition of the target pose of each link surface under multiple manipulator poses,so its visibility is affected by narrow field of view angle,abnormal orientation of the target plane and the occlusion of the manipulator body at the same time,which increases the construction difficulty of measuring the pose.Therefore,a model-free calibration pose selection method based on multi-object vision was proposed.According to the parameters and contour size of the links of the manipulator,the oriented bounding box trees that fitted the equivalent model was generated,and the fixed connection position of the target was set according to the manipulator′s shape characteristics.On the basis,the criterions of the target inclusion,the angle between the target plane normal direction and the line-of-sight,and the interference between the surrounding box and the target plane cone were established respectively,which were fused into the multi-target composite visibility constraint conditions.The algorithm can be used to select the pose of the manipulator with all targets visible under the established working trajectory.Finally,an experimental bench was set up to verify the effectiveness of the proposed method.The experiment results show that the method can accurately judge the occlusion relationship between the arm and the target surface,and the target has good visibility at the selected pose,which meets the requirements of the real-time model-free calibration method for the simultaneous acquisition of multi-target poses of the manipulator.