In many automation application scenarios, such as assembly and sorting processes, the use of industrial robots is an important part of improving production quality and productivity. During the working processes, safety is one of the primary prerequisites to be considered.[ In this paper, a safety warning system based on 3D vision for robots is proposed. First, the system uses 3D cameras to perform high-precision 3D reconstruction of the monitoring scene and fuse multiple point clouds; secondly, the human key points are extracted and the safety distance between the human and the robot arm is judged according to the key|points; finally,the calculation determines whether the human and the robot are at the set safety distance and controls the working state of the robot arm. Experimental results show that the proposed safety warning system is able to work within a large field of view.