Extracting Tree Biophysical Parameters from Horizontally Displaced Stereoscopic Images

Abstract

Forests are very important because they provide food, fuel, and medicine to many populations in the world. They also aid in climate change mitigation since they are the largest terrestrial carbon sink. Because of the many benefits drawn from forests, it is essential to collect and store measurement data to help monitor them. The most commonly measured tree biophysical parameters in forest inventory are the diameter at breast height (DBH) and tree height (TH). Manual techniques used for forest inventory are slow and labour-intensive because tools often have to come into contact with trees, making it necessary to deploy more personnel to speed up the process. Many non-contact techniques proposed by researchers are expensive, computationally heavy, require extensive training or exhibit low accuracy. This study presents computational geometric algorithms for estimating the breast height (BH) location and extracting the DBH, TH and crown diameter (CD) from disparity images derived from stereoscopic images. 105 image pairs of trunks belonging to 20 trees and 10 image pairs of full trees were acquired in 30 minutes. Images for locating the BH were taken at 5m, 6m, 7m, 8m and 9m, those for estimating the DBH at 5m and 8m, and those for estimating TH and CD at arbitrary distances. The results indicated RMSEs of 1.02cm (R ²=0.9929) and 0.94cm (R ²=0.9946) in DBH estimation at 5m and 8m respectively, MAE of 3.15cm in BH location, and RMSEs of 19.32cm (R ²=0.9849) and 31.93cm (R ²=0.9599) in TH and CD estimation respectively. This implies that the technique can be used for DBH and TH estimation with very good accuracy. CD and TH estimation can be done only for short trees (< 6m height), albeit with a slightly lower performance for CD.

Keywords: Forest inventory, stereoscopic vision, diameter at breast height, crown Diameter, tree height, disparity image