Crops As Time-Invariant Keypoints
M. Bos (TU Delft - Electrical Engineering, Mathematics and Computer Science)
NV Budko – Mentor (TU Delft - Numerical Analysis)
E. Verhoeff – Mentor (VanBoven Drones B.V.)
C. Vuik – Graduation committee member (TU Delft - Numerical Analysis)
R. F. Remis – Graduation committee member (TU Delft - Signal Processing Systems)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
In this paper, a method is proposed to automatically correct misalignment of orthophotos in time-series caused by an inaccurate geotransform. The proposed method relies on common literature concepts such as keypoint identification, keypoint matching, and model fitting using random sample consensus (RANSAC). Traditional keypoint identification methods such as the scale invariant feature transform (SIFT) are not suited for this problem as no real scale- or rotation-invariance is required, instead, time-invariance is required. To achieve this, crops are suggested as keypoints, and two different keypoint descriptors are put forth. The first descriptor is based on the shape and size of the crops, while the alternative descriptor is based on the planting pattern of crops. The method, and both descriptors, generate promising results for certain scenario's. However, in later growth stages performance drops significantly as the identification of crops -required beforehand- becomes troublesome due to dense growth.