Active camera repositioning method based on three-dimensional point cloud alignment

Abstract

The invention discloses an active camera repositioning method based on three-dimensional point cloud alignment, and the method is completed through the cooperation of three parallel threads, achievesthe three-dimensional reconstruction of a scene at an initialization stage through an existing visual SLAM method, and obtains the relative pose relation between a platform and a camera, and the realscale of three-dimensional reconstruction. After that, three parallel threads are constructed: 1) a visual three-dimensional reconstruction algorithm (VSE) thread: obtaining the three-dimensional point cloud of the current scene and the pose of the current camera; 2) a point cloud alignment thread: completing alignment of the reference point cloud and the current point cloud, and obtaining scale transformation and coordinate transformation between the two point clouds in real time; 3) a camera adjusting thread: calculating and executing the adjustment amount of the platform; according to the invention, the dependence of the existing relocation method on the illumination consistency of the previous observation scene (especially the outdoor environment) is effectively reduced, and meanwhile,the relocation time is effectively reduced, so that the method can be successfully applied to a tiny change detection task in a real scene.

Description

technical field [0001] The invention belongs to the fields of artificial intelligence and computer vision, relates to active vision technology, and specifically relates to an active camera relocation method based on three-dimensional point cloud alignment. Background technique [0002] Active camera repositioning aims to physically and truly restore the camera's six-degree-of-freedom pose to the same as when shooting the reference image. It plays an important role in environmental monitoring, preventive protection of historical and cultural heritage, and detection of small changes. It is an active An important application of vision technology. The active camera relocalization process includes the relative pose estimation of the camera and the dynamic adjustment of the camera, and the adjustment of the camera is completed by the robot platform. [0003] State-of-the-art active camera relocalization algorithms have achieved great success in the task of small change detection ...

AI Technical Summary

Problems solved by technology

[0004] However, when the lighting conditions (direction and intensity) of previous observations are different, the monitoring results are not satisfactory

Significant differences in illumination will change the appearance of the scene (especially the ubiquitous three-dimensional structure scene), and then the feature point descriptors in the image involving pose estimation will change, resulting in the failure of camera pose estimation

In addition, if the background (non-monitoring object area) in the observation scene changes greatly, for example, the vegetation near the monitored ancient building may change drastically in different seasons (even structural changes), which will significantly increase the probability of mis-matching feature points in the image. Quantity, which seriously affects the accuracy of relocation results

The above two situations that are common in actual tasks will seriously damage the accuracy of active camera relocation, resulting in unreliable actual monitoring results, so that it cannot support camera relocation operations in outdoor uncontrollable environmental conditions

Images