الفهرس 
AbstractOnly 14 pages are availabe for public view
 |  | from | 161 |  |  |
| | | from | 161 | | |
Abstract
One of the most efficient methods for recognizing 3D objects can be achieved through building complete, accurate, detailed, and realistic 3D models of these objects from still images. To reconstruct a 3D model of an object, firstly features are extracted from images of that object taken from different viewpoints, then these features are matched and after finding correspondences, 3D information can be obtained. In general, 3D object recognition techniques are mainly depending on the reconstruction of a 3D model of an object; which is considered a difficult task if only a few numbers of images of that object are available. To overcome this problem, several methods can be used to create new images from the original dataset such as horizontal shift, rotation, vertical shift, and composite shift. Applying feature extraction and matching between these images, a 3D model can be reconstructed. On the other hand, when two or more images of an object are available, multiple 3D models can be obtained by performing multistage features extraction and matching on these images. There are several algorithms that can be used for feature extraction and matching. In this work, four algorithms; (i) Scale Invariant Feature Transform (SIFT), (ii) Kanade-Lucas-Tomasi (KLT), (iii) Corner Detector (CD), and (iv) Speeded Up Robust Features (SURF) will be used. Performing feature extraction and matching by different algorithms results in building different 3D models. Applying different data fusion techniques between different 3D models can produce a more accurate 3D model. In this work, four data fusion techniques; Addition, Simple Average, Principal Component Analysis (PCA), and Discerete Wavelet Transform (DWT) will be tested. Experimental results showed that the presneted proposed methods provide a greate improvement on the performance metrics in the two phases; reconstruction and recognition, respectively.