Research On Visual Odometry/SINS Integrated Navigation System

As a kind of autonomous navigation system that do not rely on outside information, SINS is widely used in robot navigation,ship navigation, aircraft guidance, underwater navigation field due to its high reliability, low cost, small volume. SINS/GPS integrated navigation system overcomes the shortcomings of SINS error accumulating with time by introduction GPS's position information and velocity information. But GPS signal is susceptible to electromagnetic interference,easy to be shielded, controlled by others, and can not achieve the completely independent. With navigation system's request for autonomy,crypticity being higher and higher both in military fields and civilian fields, visual odometry/SINS integrated navigation system as a completely independent navigation system has gradually become an important research direction of navigation technology. This paper researches binocular visual odometry/ SINS integrated navigation system in application of robot, the main research contents are as follows:Before the study of binocular vision odometry/SINS integrated navigation system, the paper introduces the basic principle of SINS and visual odometry. For SINS, mainly introduces the definition of common coordinate system, derives inertial navigation basic equations, establishes SINS error model on moving base. For binocular vision odometry, introduces the common coordinate systems and the transformation between coordinate systems, derives mathematical model of binocular visual odometry, details the camera calibration, image correction, feature points extraction and matching, motion estimation and so on.Adopting visual's position and attitude as measurement, kalman filter model is established by the INS error model and visual/SINS integrated navigation system simulation analysis is carried out. Simulation results show that when the system model is accurately known, visual/SINS integrated navigation system has the same parameter precision with GPS/SINS integrated navigation system and its attitude error converges faster; when the system in the presence of random bias and statistical characteristics of noise not accurately known, the navigation accuracy of visual/SINS integrated navigation system decrease. It shows that the unknown system noise and random bias makes the filtering performance of KF decrease, leading to navigation accuracy decrease.To solve the above problems, put forward a kind of visual/SINS integrated navigation system based on Adaptive Two-stage Kalman Filter (ATKF). This paper improves the traditional ATKF according to the method of solving genetic factor. The improved ATKF can be respectively suitable for the two cases, namely system noise statistical model is not accurately known and measurement noise statistical model is not accurately known. The two cases respectively correspond to Tuning Pk ATKF and Tuning Rk ATKF and both of them have unified filtering framework. Simulation results show that the convergence precision of visual / SINS integrated navigation system based on improved ATKF higher than traditional algorithm and filter divergence caused by inaccurate model is effectively solved.The different data update cycle of SINS and binocular vision odometry and the presence of output delay of binocular vision odometry are the main reason leading two subsystems data asynchronous. For different data update cycle problem, this paper use Multi-rate Kalman Filter as the solution and use the smaller SINS cycle as filter cycle. when visual information not existence, the system only updates time;when visual information not existence, the system updates time and measurement information. Simulation results show that Multi-rate Kalman Filter improves the filter precision and data utilization. For output delay problem, this paper extrapolates the lagging visual information to filtering moment using second-order extrapolation method and uses the extrapolated data for filtering. Simulation results show that second-order extrapolation method has a higher precision and can effectively solve visual output delay problem.