Cabin robot can provide assistance for astronauts in the cabin. On the one hand, it can be used to alarm the smoke and other contingencies in the cabin to ensure personal safety. On the other hand, when astronauts are lonely, it can monitor the mental/physical health of the astronauts in the cabin. For this purpose, real-time visual tracking and positioning of the second generation Astronaut Assistant Robot (AAR-2) is carried out. Because of the limitation of the cabin space, with the movements of the robot, sometimes only the local/partial images of targets can be captured by camera. So how to carry out the tracking and pose calculation through those partial images is a challenge. The paper proposes an integration method with a target feature template and a local search algorithm to track a moving target in real time, where the position and attitude of the camera are calculated using the PnP algorithm. Camera calibration and moving target tracking are expounded in detail. Furthermore, the accuracy is verified by experimental results. It turns out that the visual positioning system meets the precision requirement in real time.