Field-of-View Expanding For Tree Images Capturing Based On Image Mosaic

Presently forest plant diseases and insect pests have highly restricted afforestation and ecological environment construction. The traditional chemical application treatment, however, has very low efficiency. Therefore it becomes necessary to develop precision pesticide application methods based on machine vision technique. Aiming at solving the problem in the precision pesticide application technique that it is unable to capture the whole tree image when a tree is too tall and the field-of-view is too narrow, parts of tree images were combined into an integrated tree image with the technique of image mosaicing, and the depth information was acquired during the process which provided a basis for adjusting the coverage and range of pesticide spraying accurately.Firstly, the characteristics of images of trees were studied and a set of methods were explored which were applicable to trees image mosaicing including segmenting background and tree, searthing for featured regions, determining the template, searching for the matching point, calculating the mosaicing parameters, preprocessing the original tree images, mosaicing the initial tree images, processing the mosaicing seam, completing tree images mosaicing and so on.Secondly, an analysis system was developed based on the single-group of tree images mosaicing. The system divided the single-group of tree images mosaicing processing into several small steps and it could run on single-step operating and display the intermediate states. During the mosaicing process the different algorithms and parameters could also be chosen, which provided a platform for analyzing the results and optimizing the algorithms and laid the foundation for the establishment of real-time mosaicing system.Finally, a real-time dynamic tree images sequence mosaicing system was developed and the real-time tree images mosaicing hardware platform was built up, which contained a computer, a set of digital CCD camera (with two cameras), a framework for the cameras ,a 1394 card and simulated trees. The real-time process from tree images capturing to mosaicing could be realized by the system. Simultaneously, based on the basic principle of binocular vision through two cameras on trees, the spatial relations were studied between cameras and trees and thus a geometric model was established. By utilizing such parameters as the camera field-of-view, the angle between optical axis, the distance of two focus and so forth, and the tree depth information could be obtained through combining with the acquired tree image mosaicing parameters.The research achievement in this paper can provide valuable experiences both theoretically and practically for improving implementation of precision pesticide application.