In recent years, mobile operating robots have been widely used in industry and service industries due to their flexibility and efficiency. However, the acquisition of position information of mobile robots in a collaborative environment is still challenging. In this paper, a multi-stage robotic arm 3D pose estimation method based on RGB images is proposed. This method aims to obtain the 3D pose of the robot arm of the mobile robot through an RGB image, thereby reducing the positioning accuracy requirements and communication construction of the mobile robot in the cooperation scene. Our method has three main contributions. First, we provide a method for constructing an automatically labeled dataset. Then we propose a key-point estimation network with two branches to extract key-point positions of the manipulator and their confidences. Finally, we obtain the optimized 3D pose of the manipulator by combining the robot and the camera model. The experimental results show that the error of the spatial pose estimated by the method in this paper is less than 20mm. Experimental demonstration video will be posted on https://youtu.be/1fohZRQzMc4.