Publikation
Feature-Based Image Selection (FBIS) for Enhancing 3D Reconstruction in Agriculture
Jomana Abdelmoaty; Zoltán Istenes
In: Proceedings of the 12th International Conference on Mechatronics and Robotics Engineering (ICMRE 2026). International Conference on Mechatronics and Robotics Engineering (ICMRE-2026), March 2-4, Oldenburg, Germany, IEEE Xplore, 2026.
Zusammenfassung
The future of technology increasingly relies on artificial intelligence systems, which must be capable of perceiving visual scenes like humans to interpret and understand their surroundings, making 3D reconstruction a fundamental component of this capability. This paper focuses on 3D reconstruction of an apple plantation for monitoring purposes using a mobile robot equipped with an RGB-D RealSense D435 camera. It presents a comparative analysis of two widely used reconstruction tools, COLMAP and Open3D, which utilise different input data types: RGB for COLMAP and RGB-D for Open3D, to identify the most suitable approach for outdoor environments exposed to sunlight. To reduce processing overhead and improve reconstruction efficiency, a Feature-Based Image Selection (FBIS) algorithm is proposed. FBIS leverages feature extraction techniques to eliminate redundant images that lack novel features. By evaluating the performance of COLMAP and Open3D, this work offers insights into their suitability for field conditions and proposes an effective strategy to optimise 3D reconstruction in agricultural settings. The apple plantation datasets are available at https://doi.org/10.5281/zenodo.17183457.