3D reconstruction of logs from board images

Abstract

Sawmills nowadays are heavily automated and especially sawing process optimisation could improve overall yield of sawn timber. The main focus of this thesis work is to create a method to reconstruct a 3D-model of a log based on RGB images of boards. The board images can be considered as cross-sections of the original log and can be used to obtain information about the inner structures of the log. By using the board surface knots, an approximate knot growth model can be fitted to the detected and clustered knot edges. This information can be then extrapolated to see how the knots could appear on the log surface. A proposed approach can be beneficial for getting better understanding of knot growth and thus developing better log modelling systems for sawmills. The implemented pipeline acts as a groundwork for future improvements as e.g. knot clustering has much room to improve. Current results are more promising with lower quality boards because the knots are more defined and more numerous. With higher quality boards the results are generally worse and also there is a larger variance within the results than with lower quality boards.