Stereo Visual-Inertial Odometry Estimation Approach

Positioning technology is a key technology in the domains of moving robots, vehicle navigation, ocean sailing, and aviation etc. Positioning technology includes absolute positioning technology and relative positioning technology. Relative positioning technology is significantly important, especially in situations where absolute positioning technology is invalid.Stereo visual odometry and inertial navigation are both important relative positioning technology. The former is very precise, but depends on the quality of camera images which suffers from blurring and illumination variation etc. The later captures stable sensor data, but suffers from large error accumulation along time. Fusing them will lead to better odometry method which can combine the high precision of stereo visual odometry and the stability of inertial navigation.This thesis undertakes research on stereo visual odometry and inertial navigation, and then focuses its mind on the fusion of them. The main contributions are as follows:1. Propose a three kalman filters ensemble visual-inertial fusion algorithm. Three kalman filters each corresponding to a different data rate can avoid the errors of velocity estimation caused by visual key frames.2. Introduce the indirect kalman filter to deal with the orientation fusion. Indirect kalman filter fuses orientation in linear error space other than non-linear space of orientation. Introduce Modified Rodrigues Parameters to transform the state of the filter avoids the approximation of the state in the equation.3. Prove the stability of the filters theoretically.4. Build an environment perception system which can function real-time odometry estimation.