A Space Non-cooperative Target Relative Navigation Approach Using Monocular Vision And Inertial Measurement Unit

With the development of on-orbit services and space attack and defense,the demand for spacecraft relative navigation,especially spacecraft autonomous relative navigation,is increasing.The camera is a commonly used sensor in space relative navigation,which is small,light,cheap and efficient.However,the visual navigation accuracy is limited because the limitation of the size of sensor.Combining the measurement of camera with the measurement of the inertial measurement unit(IMU)and using them in spacecraft relative navigation can improve the navigation accuracy.Navigation need not only measurement but also optimization.The Kalman filter has low accuracy and poor robustness,and the least square method which usually run offline because its large amount of calculation cannot optimize sequential data.So the paper uses sliding window filter which has high accuracy with limited computation.The paper can be divided into three parts.For non-cooperative targets whose model is known,scale information can be obtained from the model,and the position of the satellite's center of mass is known.Since the refresh rate of the IMU is much higher than that of the camera,a pre-integration method is proposed to accumulate the IMU data between the two frames of the camera as a single frame of data for obtaining an accurate motion state.At the same time,mathematical simulation can show the changes in navigation accuracy and calculation when sliding windows has different amounts of state variables.More data is add in sliding windows,higher accuracy can get by the optimization;on the contrary,less data is add in sliding windows,lower accuracy can get by the optimization.For non-cooperative targets which only part of the model is known,the scale information still can be obtained,but the position of the center of mass is unobservable.Since the center of mass is related to the estimation of the motion state of the noncooperative target,it's necessary to estimate the position of the center of mass in the same time.Although the position of the center of mass is constant,it can be optimized in the sliding window at the same time as estimating the motion state.Better measurement accuracy can be obtained.Except for the above-mentioned non-cooperative targets with model information,most of the non-cooperative targets are completely unknown about the model,so it's needed to rebuild the target model while estimating the motion state.In order to reduce the calculation,using only a single parameter,the inverse depth,to refer to the position of feature.On the premise of ensuring the rebuilding accuracy,reducing three parameters to one can reduce calculation.This method uses only a monocular camera and IMU to complete the functions of relative navigation and three-dimensional rebuilding,and is a feasible spatial SLAM method.