Date of Award
2020
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Computer Science
Committee Chair
Ramazan S. Aygun
Committee Member
Farbod Fahimi
Committee Member
Letha H. Etzkorn
Committee Member
Tathagata Mukherjee
Committee Member
Haeyoung Chung
Subject(s)
Robot vision, Robots--Control systems
Abstract
This dissertation proposes a new method for real time vision-based localization in the full 360-degree area around a 3D reference object. In the past, Mirage pose estimation was developed for multi-camera systems. To achieve 360-degree localization with a single camera, we introduce the Mirage-S pose estimation method that uses a single camera given a 3D reference object as input. Mirage-S finds the solution by using an iterative numerical method with non-linear equations. This dissertation splits the localization problem into three parts: 360-degree 3D object recognition, 360-degree pose estimation, and 360-degree localization application. First, 360-degree 3D object recognition is required to recognize a reference 3D object from all possible 360 degree locations. Second, we use Mirage-S for 360-degree pose estimation to calculate the pose of a robot at all possible 360 degree locations around the reference 3D object. Third, a 360-degree localization application is integrated with a closed-loop real-time trajectory tracking controller. This application expects the mobile robot to move on a desired path covering the whole 360-degree area around the 3D reference object. The outcome of the simulations and real experiments using noise and noise-free information are presented to evaluate the performance of Mirage-S. The results are compared to eight well-known Perspective-n-Point(PnP) methods and Mirage with multiple camera systems. Furthermore, the comparison of the real experimental results between Mirage-S and PoseNet, a learning-based architecture that regressed the absolute pose using a deep convolutional neural network is presented. The successful results of the simulations and real experiments of trajectory tracking are provided.
Recommended Citation
Singhirunnusorn, Khomsun, "A single camera 360-degree real time vision-based localization method with application to mobile robot trajectory tracking" (2020). Dissertations. 203.
https://louis.uah.edu/uah-dissertations/203