2-D delineation of individual citrus trees from UAV-based dense photogrammetric surface models. Issue 6 (3rd June 2018)

• Record Type: Journal Article
• Authors: Ok, Ali Ozgun; Ozdarici-Ok, Asli
• Is Part Of: International journal of digital earth. Volume 11:Issue 6(2018)
• Page Start: 583
• Page End: 608
• Publication Date: 2018-06-03
• Subjects: Orientation-based radial symmetry -- active contour -- citrus trees -- UAV digital surface model; Geographic information systems -- Periodicals; Sustainable development -- Information technology -- Periodicals; Social planning -- Information technology -- Periodicals; 910.285
• DOI: 10.1080/17538947.2017.1337820
• Languages: English
• ISSNs: 1753-8947

Abstract

ABSTRACT: One of the challenges of remote sensing and computer vision lies in the three-dimensional (3-D) reconstruction of individual trees by using automated methods through very high-resolution (VHR) data sets. However, a successful and complete 3-D reconstruction relies on precise delineation of the trees in two dimensions. In this paper, we present an original approach to detect and delineate citrus trees using unmanned aerial vehicles based on photogrammetric digital surface models (DSMs). The symmetry of the citrus trees in a DSM is handled by an orientation-based radial symmetry transform which is computed in a unique way. Next, we propose an efficient strategy to accurately build influence regions of each tree, and then we delineate individual citrus trees through active contours by taking into account the influence region of each canopy. We also present two efficient strategies to filter out erroneously detected canopy regions without having any height thresholds. Experiments are carried out on eight test DSMs composed of different types of citrus orchards with varying densities and canopy sizes. Extensive comparisons to the state-of-the-art approaches reveal that our proposed approach provides superior detection and delineation performances through supporting a nice balance between precision and recall measures.