STEM HEIGHT AND LEAF LENGTH ESTIMATION FOR GROWING RATE CALCULATION USING 3D RECONSTRUCTION FROM MULTI-VIEW IMAGES

Abstract

Agriculture is one of the key sectors contributing to Vietnam's economic development, providing employment, and income for the population, and ensuring
food security. However, global climate change and population growth currently pose significant challenges to agricultural production. To address these
challenges, new techniques have been proposed to enhance crop quality and increase yields. This study proposes a novel method for estimating plant height
and leaf length using multi-view images. Key features are extracted from the collected data, and correlations between images are identified, allowing for the
determination of sparse point clouds. Next, the position and angle of the images are initialized in three-dimensional. Using orthogonal projection, dense point
clouds of the object are generated. The plant's skeleton is extracted from the dense point clouds using the Laplace contraction method. Then, different parts of
the plant, such as leaves and stems, are segmented for length calculation. In this study, five maize plants were grown for 20 days, with images taken at five-
time points. The proposed method reached an accuracy of 95%, RMSE is 0.311, and R2
is 0.82. These promising results will promote the application of new
technologies in agriculture to improve crop quality and yield.