A Portable Computer Vision System Forin-Situ Leaf Area Measurement Outdoors

• Main Author: Yau, Weng Kuan
• Format: Final Year Project / Dissertation / Thesis
• Published: 2021
• Subjects: TA Engineering (General). Civil engineering (General)
• Online Access: http://eprints.utar.edu.my/4604/1/Yau_Weng_Kuan.pdf; http://eprints.utar.edu.my/4604/

Plant phenotyping is a research area concerned with the quantitative measurement of a plant’s structural and functional properties. Most common methods for measuring a plant’s individual leaf surface area are laborious and stressful to the plant. Therefore, there is a push to utilize depth sensors as a contactless, non-destructive and in-situ method for measuring individual leaves surface area. Consumer RGB-D sensors like the Asus Xtion Pro Live, Microsoft Kinect v2 and the Intel Realsense R200 each employ different depth sensing technologies. We first compare all three sensors for capturing 3D surface data of objects outdoors under strong sunlight NIR interference. The Kinect v2 was proven to be the most suitable sensor in our use-case scenario for capturing 3D surface data of plants outdoors for the purpose of measuring individual leaves area. In order to measure the surface area of individual leaves, each leaf of interest needs to be segmented out from the captured 2.5D point cloud of plants in a complicated natural scene. HDBSCAN clustering was employed to cluster-segment out individual leaves. Performance of leaf segmentation was measured by evaluating the nearest 10 (max) clusters of data obtained into three categories, individual leaves, under-segmented and over-segmented. Probability of segmenting individual leaves differs from plant to plant, ranging from a low of 0.7178 to a high of 0.8975. The surface area of all individual nonoccluded leaves obtained via the segmentation method was calculated and compared to its ground truth. The calculated individual leaf surface areas R2 recorded ranged from i0.792 to 0.911 with respect to its best fit regression line while the RMSE range from 4.9482 to 14.4941 cm2. The proposed system and method was able to segment individual leaves from dense foliage for the purpose of measuring its surface area.