| As compared with traditional aerial and space platforms, the UAV platform offers greaterflexibility, shorter response time, lower cost and is able to generate very high resolution imagedata. The UAV remote sensing imagery can be widely used in large scale mapping of small areacoverage. The new style centimeter-level spatial resolution remote sensing imagery acquired byUAV low-altitude remote sensing system is an important expansion of conventional remotesensing data source. Ground objects extraction from remote sensing imagery has always been achallenging task, this thesis aims at evaluating the segmentation of UAV remote sensing imageryand developing an object-based extraction method for ground objects. The main works are listedas follows:1) The existing image segmentation algorithms are described. Different methods proposedso far for segmentation evaluation are studied. Four classic segmentation algorithms(optimal edge detector, watershed segmentation, otsu multi-thresholding,region-growing) are applied on three UAV images (containing different ground objects).In order to compare the algorithms, an evaluation of each algorithm is carried out withempirical discrepancy evaluation methods. This evaluation is carried out with a visualsegmentation of UAV remote sensing images;2) Theory of object-oriented image processing is studied. A decision tree was used todetermine the optimal texture features for each segmentation scale. The experimentalresults demonstrate that the combination of decision tree method and object-orientedimage processing can cut down the workload of ground objects extraction from UAVremote sensing imagery;3) An object-oriented vehicle extraction method from UAV remote sensing imagery isproposed. A decision tree is developed by training data, combined with assistantcriterions, the vehicles in UAV remote sensing imagery are detected and classified.Vehicle motion information extraction from image sequence is also described. |
PDF Full Text Request