The object transfer between human and robot is a hot and significant research direction in human-robot interaction. We propose a method for transferring object from a robotic arm to a human. Based on depth filter binarization and object geometry, the robotic arm can grasp the object from the table without knowing the type of object in advance. Different from the methods based on the detection of the body or arm in previous studies, based on the identification of the key points of hand and the singular value decomposition(svd), the hand posture can be determined. So we can drive the robot to pass object to the human in the appropriate posture. Through 200 experiments on 10 kinds of objects, the effectiveness of the method is proved, and the grasp process and the transfer process of the object have a high success rate. This method can be applied in industry and life fields.