Inertial monocular visual odometry based on RUPF algorithm

The accurate autonomous positioning performance of the unmanned vehicle is an important prerequisite for autonomous navigation. The combination of visual and inertial sensors is a cheap, compact and complementary solution. In this paper, in order to improve the positioning accuracy of the monocular vision/IMU integrated system, an visual inertial odometry based on the Random sampling consistency of the Unscented Kalman Particle Filter (RUPF) was proattituded. With respect to feature point tracking, the method of optical flow tracking is used to reduce the time required for feature matching. At the same time, a random sampling consistency algorithm is used to eliminate the mis-matching feature points in the three-view feature point tracking process, and then the Epipolar geometry and the three constraints formed by the focus tensor is integrated into the observation equations, and unconstrained particle filtering is used to fuse the IMU and monocular visual information. The fusion algorithm is verified by the KITTI sports car data set. The experimental results show that the visual inertial odometry based on the RUPF filtering algorithm has accurate positioning accuracy, the final positioning error is controlled at around 0.19%.