A monocular vision odometer method adopting deep learning and mixed pose estimation

Abstract

The invention discloses a monocular vision odometer method adopting deep learning and mixed pose estimation. The monocular vision odometer method comprises the following steps: estimating an optical flow field between continuous images by utilizing a deep learning neural network, and extracting key point matching pairs from the optical flow field; taking key point matching pairs as input, and according to a 2d-2d pose estimation principle, calculating a rotation matrix and a translation vector preliminarily by using an epipolar geometry method. A monocular image depth field is estimated by using a deep neural network, a geometric theory triangulation method is combined, and the depth field serves as a reference value, thus calculating by using an RANSAC algorithm to obtain an absolute scale, and converting the pose from a normalized coordinate system to a real coordinate system; and when the 2d-2d pose estimation fails or the absolute scale estimation fails, performing pose estimationby using a PnP algorithm by using a 3d-2d pose estimation principle. According to the invention, accurate pose estimation and absolute scale estimation can be obtained, the robustness is good, and thecamera track can be well reproduced in different scene environments.

Description

technical field [0001] The invention relates to a monocular visual odometer method, in particular to a monocular visual odometer method using deep learning and hybrid pose estimation, and belongs to the technical field of simultaneous localization and mapping (SLAM). Background technique [0002] At the same time, positioning and map construction are mainly used in the fields of robots, drones, unmanned driving, augmented reality, virtual reality, etc. Localization and mapping problems while moving in unknown environments. As one of the core components of simultaneous positioning and map construction, visual odometry can locate the position of the robot itself in the environment and estimate the relative motion state information of 6 degrees of freedom, including displacement information of 3 degrees of freedom and rotation of 3 degrees of freedom Information, and then calculate the absolute position information through the relative motion information, and then reproduce th...

AI Technical Summary

Problems solved by technology

Similar to the first method, this method still only relies on the neural network to estimate the pose state, without taking advantage of the advantages of geometric theory in pose estimation. The generalization ability of this method needs to be improved.

Images