Georgia Institute of Technology
Pixel Inverse Depth Parameterization, Theory and Application
Abstract
dc:description.abstractThis research seeks to improve upon inverse depth parameterization (IDP), a standard method used to represent spatial coordinates of feature points in map estimation. If the original measurement noise is Gaussian, the IDP can linearize the estimation process and maintain the Gaussian distribution. However, the initialization of IDP is heuristic, not analytic, due to the coordination characteristic. This research proposes a new method, termed pixel inverse depth parameterization (PIDP), which enables an analytic initialization. The proposed technology makes the initialization possible to establish a correct connection between initialization and update probabilistically. This research also introduces two applications of PIDP. First, the PIDP property of being able to initialize leads to a proposed method for estimating spatial points with an array of multiple cameras, including an algorithm and an accompanying optimization method. Although there is a method of estimating a spatial location with a stereo camera, there are few methods for combining and estimating by using more than two cameras – which is enabled by less expensive cameras and processors. This method does not require iterative calculations. The proposed method makes estimation possible with one regression. As a second application, PIDP is implemented and tested in an error-state Kalman filter for a robust state estimation method. Then, a comparison between PIDP and IDP is performed by simulation.
Degree
thesis:*- Level thesis:degree_level
- Doctoral
- Department dc:contributor.department
- Aerospace Engineering
- Grantor dc:publisher
- Georgia Institute of Technology
- Year dc:date.issued
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Watanabe, Toshinobu
- Advisors dc:contributor.advisor
-
- Johnson, Eric N.
- Prasad, Jonnalagadda V. R.
- Committee members dc:contributor.committeemember
-
- Feron, Eric M.
- Zhou, Hao-Min
- Vela, Patricio Antonio
Subjects
dc:subject × 8Rights
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/71958