Three-dimensional citrus grove reconstruction based on empirical measurement.

摘要

My study contributes to the development of unmanned and automated techniques associated with two applications for citrus groves: ground-based scouting autonomous vehicles and harvesting robot manipulators. The duty of ground-based scouting autonomous vehicles is to collect global information, threading along alleyways in a grove, and harvesting robot systems explore and map local fruit positions on a canopy using vision systems. Two major techniques considered in my study are a global image mosaicing technique for the ground-based scouting autonomous vehicle and a local 3D reconstruction for the harvesting robot manipulator. Vision-based techniques follow common procedures related to features: feature detection, feature tracking and feature matching.;Since it must be assumed that there are no artificial landmarks in canopy scenes, it is necessary to extract features from leaves. A leaf detection algorithm was developed by combinations of morphological operations designed to look for the center of the leaf. Leaf detection can preserve the viability of dominant interest points by means of segmenting individual. To be robust against the disturbance and disappearance of interest points, an active mesh method is introduced. It was originally designed to detect shapes of objects, however, the method is used in tracking with an assumption that objects are not deformable during consecutive frames. A multilayered active tree was invented to be more robust to noises by managing interest points hierarchically. The multilayered active tree is not only stable, but it determines the dominant features. Some of the matched sets are used in the estimation of optical flow and reconstruction. For the horizontal movement, a parallax which causes discords between images occurs. A sequential image mosaicing was implemented by cropping and stitching the overlapped area between images. Difficulties of an 8 point algorithm were pointed out and a Plucker coordinates system which provides simpler and intuitive computations was suggested for the 3D canopy reconstruction. An alternative way to reconstruct canopies was also conducted through a LADAR and a GPS receiver.