Binocular VIO implementation method based on variational Bayesian adaptive algorithm

Abstract

The invention provides a binocular VIO implementation method based on a variational Bayesian adaptive algorithm, and the method comprises the steps: obtaining an image through a binocular camera, carrying out the feature point extraction of the image, and adding the extracted feature points into a globally maintained map container; carrying out IMU state prediction through the IMU data; when a newframe of image is received, adding the frame of image into the state vector to augment the state vector and the state covariance; judging whether feature points exist or whether the camera needs to be deleted, and if yes, performing filtering fusion; if yes, performing UT transformation based on an observation model of the binocular camera, and calculating a Jacobian matrix corresponding to the observation model; superposing the plurality of Jacobian matrixes, and performing null-space projection to obtain a final standard observation equation; and applying variational Bayesian estimation tothe obtained standard observation equation, and updating the state of the VIO system. The time-varying situation of system observation noise can be well processed, and the robustness is improved whilethe precision is improved.

Description

technical field [0001] The present invention relates to the field of robot positioning, in particular to a method for implementing binocular visual inertial odometer (Visual Inertial Odometry, VIO) based on variational Bayesian (variational Bayesian, VB) adaptive nonlinear filtering. Background technique [0002] The problem of simultaneous localization and mapping (SLAM) is a crucial part for realizing a fully autonomous robot. When we do not consider the mapping problem, the SLAM problem is simplified to the positioning problem of the odometer. The vision-based positioning algorithm is more and more popular because it can provide rich environmental information and its low cost. However, relying solely on visual information to complete robot positioning, the system is not robust enough in many cases (for example, when the texture features of the environment are not obvious, the illumination of the environment changes greatly, and the camera is blurred due to the rapid move...

AI Technical Summary

Problems solved by technology

[0004]The vast majority of VIO systems assume that the observation noise is a Gaussian distribution with known mean and variance, but in reality, the variance of the measurement noise is very large May be unknown and time-varying, in addition, there

Images