Remote sensing of vegetation structure using computer vision

Background/Question/Methods High spatial resolution measurements of vegetation structure in three-dimensions (3D) are essential for accurate estimation of vegetation biomass, carbon accounting, forestry, fire hazard evaluation and other land management and scientific applications. Light Detection and Ranging (LiDAR) is now the preferred standard for these measurements but requires bulky instruments mounted on commercial aircraft that are costly and logistically challenging to deploy. Here we present on research using open-source computer vision algorithms based on fully automated bundle adjustment photogrammetry to make high spatial resolution 3D measurements of vegetation structure and spectral characteristics using a consumer grade digital camera mounted on a hobbyist aerial platform without GPS or IMU equipment. Large sets of digital photographs (200 – 600) were acquired across two 2.25 ha test sites in Baltimore MD, USA in late summer 2009 using an off-the-shelf digital camera deployed on a kite. An open-source computer vision algorithm applied to the photographs generated 3D datasets resembling LiDAR discrete-return point clouds but with RGB spectral attributes for each point. Tree height measurements made with a laser hypsometer in the field were compared with tree canopy height models (CHMs) generated from computer vision and LiDAR point clouds.

Results/Conclusions Computer vision point clouds were geocorrected to a UTM coordinate system using a Helmert transformation by identifying features in the point cloud and corresponding GCPs in a high-resolution digital orthophotograph to an approximate horizontal accuracy of 1.2m RMSE. Mean canopy height measurements from CHMs were comparable with field measurements, with LiDAR (R2 >0.82) showing greater precision than computer vision (R2 >0.65), primarily due to challenges in delineating terrain under forest canopy during leaf-on conditions. Results confirm that computer vision can support ultra-low-cost, user-deployed high spatial resolution 3D remote sensing of vegetation structure.