Research On Key Technology Of Vision Aided Inertial Navigation Based On Multiple View Geometry

This thesis studies the filtering method of vision aided inertial navigation based on the multiple view geometry under unknown environment. The purpose is to improve the navigation speed and the performance of closed-loop navigation. The main contributions of this thesis are as follows.(1) The constraint equations of multiple view geometry are proposed from the geometrical point of view, and are proved their relationship with that proposed from the epipolar geometry point of view.Firstly, a constraint equation based on two view geometry is proposed from the geometrical point of view, and is proved that it is equivalent to that proposed from the epipolar geometry point of view; secondly, a constraint equation based on three view geometry is proposed from the geometrical point of view, and is proved that it is the necessary but not sufficient condition of that proposed from the epipolar geometry point of view.(2) An IEKF method of vision aided inertial navigation based on two view geometry is proposed.Because inertial navigation systems(INS) develop navigation errors over time due to the imperfectness of the inertial sensors, vision-aided inertial navigation has become an active research field due to its low cost and autonomous nature. EKF is the preferred method during most integrated navigation system due to its ripe theory and wide application, an IEKF method of vision aided inertial navigation based on two view geometry is developed and analyses its filtering performance through the simulation, the simulation results show that some of the vehicle states can be estimated well(such as the up position and up velocity), at the same time, the divergence of other states can be restrained(such as the east position and north position).(3) An IEKF strategy of vision aided inertial navigation based on multiple view geometry is proposed.During the navigation process, the number of the images having the common overlapping area may be two or three, firstly, a general calculation method of covariance matrix is proposed, this method is particularly suitable for solving the problem of the closed-loop navigation, and the simulation results verify the feasibility of the method; secondly, an IEKF strategy of vision aided inertial navigation based on multiple view geometry is proposed based on the real image and inertial navigation information, the experimental results shows that the position of the vehicle can be estimated well using this strategy, and this strategy can be used to solve the problem of loop navigation, this conclusion provides a theoretical basis for the design of the actual system.(4) An UKF method of vision aided inertial navigation based on three view geometry is proposed.For some navigation systems requiring high performance, the IEKF method might be not fit anymore. On the premise of the advantage of UKF and the disadvantage of EKF, a method of vision aided inertial navigation based on three view geometry is proposed. Compared with the difficulty of calculating the covariance matrix in the IEKF, there is no need to calculate the covariance matrix in the UKF, moreover, the performance of the UKF is much better than that of the IEKF.(5) An UKF strategy of vision aided inertial navigation based on multiple view geometry is proposed.Based on the real image and inertial navigation information, an UKF strategy of vision aided inertial navigation based on multiple view geometry is proposed, the experimental results shows that though the computation load of UKF is larger than that of IEKF, the position of the vehicle can be estimated better using the UKF, so the strategy of the UKF is an alternative for the navigation system taking the navigation performance as the first index and this conclusion also provides a theoretical basis for the design of the actual system.