Mobile Phone Depth Sensors for Forest Carbon Measurement

Abstract

The monitoring, reporting, and verification (MRV) of forest plots, especially their tree-trunk diameters, is critical to achieving both forest protection and reforestation goals. Today’s MRV processes are mostly manual, error-prone, and costly to carry out. In this thesis, we design and implement an app running on a smartphone equipped with a time-of-flight sensor that allows efficient, low-cost, and accurate measurement of tree trunk diameters. The core focus is on designing an algorithm to identify, segment, and compute the diameter of a target tree trunk in a depth image of a forest scene, even in the face of natural leaf and branch occlusion. The algorithm runs in real-time on the phone, allowing user interaction to improve the quality of the results. We evaluate the app in realistic settings and find that in a corpus of 55 sample tree images, it estimates trunk diameter with mean error of 7.8%. We also explore a newly released alternative to the time-of-flight sensor, Google's ARCore Depth API, which uses a depth-from-motion algorithm based on a monocular phone camera and accelerometer sensors. We conclude that this API is currently inadequate for the proposed application and offer suggestions for its improvement.