Pose estimation method for depth camera

Abstract

The invention relates to the technical field of fixed-point tracking, and discloses a pose estimation method for a depth camera, and the method comprises the steps: enabling the depth camera to be disposed on a moving mechanism for photographing, obtaining a depth image and an RBG color image, and extracting ORB feature point pairs from every two adjacent photographed images; calculating an estimated value xiP of the pose change of the camera by utilizing N ORB feature point pairs with missing depth information, and calculating an estimated value xiQ of the pose change of the camera by utilizing M ORB feature point pairs with complete depth information so as to obtain a total estimated value xi0; constructing a minimum reprojection error model fusing the ORB feature point pairs with missing depth information and the error information corresponding to the ORB feature point pairs with complete depth information, and further obtaining a corresponding jacobian matrix J; and according to the total estimated value xi0, the minimum reprojection error function and the Jacobian matrix J, calculating an optimized total estimated value xik by using a nonlinear optimization method, thereby completing estimation of camera pose change in a shooting process of two adjacent frames of images.

Description

technical field [0001] The present invention relates to the technical field of positioning and tracking, in particular to a pose estimation method for a depth camera. Background technique [0002] Due to occlusion, absorption, speckle, reflection, etc., the depth image captured by the existing Kinect camera on the market often causes missing depth areas, which makes the traditional ICP algorithm sometimes lose feature points when iterating the camera pose. As a result, the algorithm cannot converge or the error is too large. When performing point cloud registration, a large number of error points appear, resulting in a decrease in its matching point pairs. When selecting the initial pose, it may deviate from the iteration interval, which increases the number of iterations and slows the convergence speed. Convergence success rate is low. [0003] There are many research methods for Kinect's depth loss problem, the most common method is to combine color image textures to repa...

AI Technical Summary

Problems solved by technology

[0004] The invention provides a pose estimation method for a depth camera, which solves the problems that the existing method takes too long to repair the lack of depth information, the precision is not improved after optimization, and it is not suitable for real-time operations of robots.

Images