Thesis (M.Sc.) - Asian Institute of Technology, 2011

A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Computer Science, School of Engineering and Technology

Agricultural farm inspection is essent ial to maintain good crop quality. Using a robot equipped with a monocular webcam for observation could reduce time and efforts of a farmer. This thesis is a continuous research to extract fruit information from a video data. Existing 3D modeling application in AIT is analyzed to fix problems and improve accuracy of reconstructed pineapple models. Improvement in reconstruction consists of symmetric feature matching, 2D-3D point matching using multiple frames, key frame selection with used of fruit region constraint, and camera motion recovery in present of outliers. Modeling is constrained by ratio of radii for more accurate fruit models. Moreover, this t hesis adds visualization which enables users to manipulate a scene of pineapple models. Experimental results demonstrate decrease in error rate in reconstruction pipeline. The system now automatically generates parts of the farm from an image sequence.