{"title":"一种有效的无地标视觉惯性导航方法","authors":"Matthew Boler, Scott M. Martin","doi":"10.1109/PLANS53410.2023.10140080","DOIUrl":null,"url":null,"abstract":"This paper presents an efficient method of fusing visual and inertial data for navigation using the two-view tensor, also known as the essential matrix. The essential matrix encodes the up-to-scale geometric relationship between two camera poses and contains the relative rotation and direction-of-translation between them. A dense network of up-to-scale relative pose measurements is constructed by computing essential matrices between incoming images and a collection of past states which observed the same scene. As the essential matrix is computed online in many visual-inertial navigation systems (VINS) as part of the image processing front end, the proposed method introduces little computational overhead while avoiding all computations related to feature estimation. This approach can be viewed as a modification of the classical pose-graph simultaneous localization and mapping (SLAM) problem. This paper further presents a dynamic initialization method to bootstrap the velocity, orientation, and biases of an IMU. The initialization method makes use of the same modified pose-graph SLAM approach to solve for the up-to-scale relative poses of a window of camera frames before solving for orientation, velocity, and sensor biases. We validate the proposed methods by implementing them in Extended Kalman Filter (EKF) and nonlinear optimization forms and testing them on public datasets.","PeriodicalId":344794,"journal":{"name":"2023 IEEE/ION Position, Location and Navigation Symposium (PLANS)","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2023-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Essential PoseSLAM: An Efficient Landmark-Free Approach to Visual-Inertial Navigation\",\"authors\":\"Matthew Boler, Scott M. Martin\",\"doi\":\"10.1109/PLANS53410.2023.10140080\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"This paper presents an efficient method of fusing visual and inertial data for navigation using the two-view tensor, also known as the essential matrix. The essential matrix encodes the up-to-scale geometric relationship between two camera poses and contains the relative rotation and direction-of-translation between them. A dense network of up-to-scale relative pose measurements is constructed by computing essential matrices between incoming images and a collection of past states which observed the same scene. As the essential matrix is computed online in many visual-inertial navigation systems (VINS) as part of the image processing front end, the proposed method introduces little computational overhead while avoiding all computations related to feature estimation. This approach can be viewed as a modification of the classical pose-graph simultaneous localization and mapping (SLAM) problem. This paper further presents a dynamic initialization method to bootstrap the velocity, orientation, and biases of an IMU. The initialization method makes use of the same modified pose-graph SLAM approach to solve for the up-to-scale relative poses of a window of camera frames before solving for orientation, velocity, and sensor biases. We validate the proposed methods by implementing them in Extended Kalman Filter (EKF) and nonlinear optimization forms and testing them on public datasets.\",\"PeriodicalId\":344794,\"journal\":{\"name\":\"2023 IEEE/ION Position, Location and Navigation Symposium (PLANS)\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2023-04-24\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"2023 IEEE/ION Position, Location and Navigation Symposium (PLANS)\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1109/PLANS53410.2023.10140080\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"2023 IEEE/ION Position, Location and Navigation Symposium (PLANS)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/PLANS53410.2023.10140080","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Essential PoseSLAM: An Efficient Landmark-Free Approach to Visual-Inertial Navigation
This paper presents an efficient method of fusing visual and inertial data for navigation using the two-view tensor, also known as the essential matrix. The essential matrix encodes the up-to-scale geometric relationship between two camera poses and contains the relative rotation and direction-of-translation between them. A dense network of up-to-scale relative pose measurements is constructed by computing essential matrices between incoming images and a collection of past states which observed the same scene. As the essential matrix is computed online in many visual-inertial navigation systems (VINS) as part of the image processing front end, the proposed method introduces little computational overhead while avoiding all computations related to feature estimation. This approach can be viewed as a modification of the classical pose-graph simultaneous localization and mapping (SLAM) problem. This paper further presents a dynamic initialization method to bootstrap the velocity, orientation, and biases of an IMU. The initialization method makes use of the same modified pose-graph SLAM approach to solve for the up-to-scale relative poses of a window of camera frames before solving for orientation, velocity, and sensor biases. We validate the proposed methods by implementing them in Extended Kalman Filter (EKF) and nonlinear optimization forms and testing them on public datasets.