UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Individual tree detection and localization in aerial imagery Murgu, Dan

Abstract

The thesis presents and investigates the results of an automated approach for locating individual tree crowns in aerial images. The paradigm of selecting representative tree crowns in the image to be analyzed itself is used to construct the matching model, a linear manifold. The technique known as principal component analysis is used in selecting the spanning basis for the manifold's subspace, aiming at reducing the dimension of the resulting model. Once the manifold has been constructed, one can determine the Reconstruction Error, or Distance From Feature Space, DFFS for short. DFFS, in conjunction with a suitable threshold, turns out to be an effective measurement to be used for accurately locating the positions of individual trees. Experiments have been undertaken to determine this threshold automatically. The result is a model of variation for DFFS which is proposed to compensate for affine transformations in the images to be analyzed. Comparative results against the ground truth are presented. Also variations in scale and misalignment in selecting the images in the training set are considered, and how these influence the results compared to the objective ground truth. The method is simple, easy to implement and achieves good performance. It also seems to be directly applicable in other areas where one searches for objects with similar appearance, and with a vague, or hard to describe structure.

Item Media

Item Citations and Data

Rights

For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.