Research On Multi-camera Vision Localization System In Indoor Environment

Nowadays,with the increasing level of science and technology,robotics and artificial intelligence technologies have made leaps and bounds.Among them,the simultaneous localization and mapping(SLAM)technology is one of the core technologies in the robotics field.Estimating robotics' own pose and perceiving surrounding scenes in a scene without a priori information has a wide range of applications in various fields such as military,production,and life.Therefore,it has been widely studied by many scholars in recent years.In general,SLAM technology can be divided into laser SLAM and visual SLAM according to the different sensors it carries.The former is equipped with a laser sensor,which uses laser scanning and matching for tracking and localization,and the latter is equipped with a camera sensor,which uses the matching relationship between images for tracking..Due to the high cost of laser sensors and the difficulty of closing the loop,the lightweight,inexpensive and information-rich sensor of the camera is favored by researchers today,and the visual SLAM technology occupies an increasingly important position.This article uses visual SLAM technology for localization in an indoor environment.However,the indoor environment is more challenging for visual SLAM.First of all,there may be more white walls indoors.If the camera field of view is a pure white wall,this weak texture environment will lead to extraction.The walking of indoor pedestrians is likely to block the camera's field of view at a close distance and a large range,although the feature points can be extracted at this time,most of them are external points that have no effect on the calculation of the pose.The above weak texture and occlusion situation will make the camera pose calculation inaccurate or even lost.Unfortunately,this situation is very easy to occur in indoor environments and seriously affects the robustness of the visual SLAM algorithm in indoor environments.Therefore,the multi-camera solution is adopted in this paper,and the observation data of multiple cameras is used to calculate the pose of the system,so that when one of the cameras has a poor field of view,the data of the other cameras can still be used for tracking,so that the robustness of the system is greatly enhanced,and the more map point constraint also improves the tracking accuracy of the system.The main research contents of this paper are as follows:1.The establishment of a multi-camera system model.The addition of multi-camera makes the original monocular imaging model no longer applicable.The classic collinear equation is extended and applied to a multi-camera system,so that a unified projection relationship is established between the spatial points' world coordinates and the pose of the multi-camera system,the pose of each camera,and the final pixel coordinates.Also,the pose update mode of the multi-camera system is analyzed.2.External parameter calibration of multi-camera system.Since this paper uses each camera to estimate the pose of the multi-camera system,the external parameters between each camera and the multi-camera system need to be calibrated in advance.In order to make full use of the advantages of the multi-camera field of view,the multi-camera designed in this paper has no cross of fields of view between them.Based on this,this paper designs two calibration schemes that do not require infrastructure to calibrate the external parameters of multi-cameras without a common field of view.3.The application of multiple cameras in the SLAM algorithm.In order to recover the true metric scale of the map with multiple cameras without cross-of-view,the master-slave camera cross-time matching method is used to calculate the scale factor;In order to update the system pose with the observation information of each camera,this paper uses the observation of multiple cameras to optimize the camera pose and the position of the map point in the tracking and mapping thread;In order to obtain a higher closed-loop recall rate,the closed-loop method between the same camera and different cameras in the closed-loop thread is analyzed.4.System experiment and analysis.According to the principles and methods adopted in the above three chapters,experiments are designed and analyzed to verify the feasibility and effectiveness of the theoretical methods.