Scale-Invariant Render-Space Phenotyping for Time-Series Crop Monitoring with 3D Gaussian Splatting
Published in Computers and Electronics in Agriculture (under review), 2026
Render-space measurement extracts normalized plant traits (height, canopy area) from 3D Gaussian Splatting renders, where perspective projection cancels the per-session Structure-from-Motion scale ambiguity by construction. Validated on a 49-day commercial-greenhouse tomato dataset: a 2.86× improvement in cross-session stability (CV 28.0% → 9.8%), reliable pruning-event detection (p = 0.0008), and physical-reference correlation (r = 0.741), with no calibration targets required.
Recommended citation: AL Zobaer, et al. (2026). "Scale-Invariant Render-Space Phenotyping for Time-Series Crop Monitoring with 3D Gaussian Splatting." Computers and Electronics in Agriculture (under review).