Pixel-to-World Mapping for Multi-Camera Warehouse Robot Localization

Author

• Mariam Faruque SharifFollow

Term of Award

Spring 2026

Degree Name

Master of Science in Computer Science (M.S.)

Document Type and Release Option

Thesis (open access)

Copyright Statement / License for Reuse

This work is licensed under a Creative Commons Attribution 4.0 License.

Department

Department of Computer Science

Committee Chair

Ryan Florin

Committee Member 1

Andrew Allen

Committee Member 2

Vijayalakshmi Ramasamy

Abstract

This thesis presents a comprehensive framework for camera calibration and pixel-to-world coordinate mapping for multi-camera robot localization in a structured warehouse environment. The study is conducted in the APRN-ROWS laboratory, where four overhead cameras observe a planar grid of known barcode locations used as the reference coordinate system.

The proposed approach combines geometric modeling and optimization-based techniques to estimate camera parameters. Initially, camera extrinsic parameters, including position and orientation, are derived using physical measurements and geometric relationships. Principal point locations are estimated through a zoom-based alignment method, ensuring accurate correspondence between the optical axis and the world coordinate system. Intrinsic parameters are obtained through standard calibration procedures, followed by image undistortion and remapping to correct lens distortion effects.

To refine the camera pose estimation, the Perspective-n-Point (solvePnP) algorithm is applied using manually collected correspondences between known world points and their pixel projections. A comparative analysis between the geometric approach and solvePnP-based optimization is conducted to evaluate accuracy and robustness. Experimental results demonstrate that the optimized method significantly improves reprojection accuracy and consistency across multiple cameras.

The calibrated system enables reliable transformation between pixel coordinates and real-world positions, supporting accurate robot localization and enabling the development of a digital twin representation of the physical environment. The proposed framework is scalable and applicable to similar multi-camera systems in robotics, automation, and intelligent warehouse applications.

Recommended Citation

Sharif, Mariam Faruque, "Pixel-to-World Mapping for Multi-Camera Warehouse Robot Localization" (2026). Master of Science Thesis, Georgia Southern University.

Research Data and Supplementary Material

No