C. Cerrada, S. Salamanca, A. Adán, J.A. Cerrada, and J.J. Escribano (Spain)
Partial Views, Matching, Recognition, 3D Object extraction, Computer Vision
This paper presents the structure and the generation method of a database composed of virtually generated partial views of 3D real objects. The main purpose of this database is to help in the visualization of complex scenes composed of several objects where only partial range data information of each object is available. The database stores the principal components of the different partial views of an object, called Virtual Partial Views (VPV), which are virtually generated off line. An algorithm to estimate the location of a given partial view with respect to the complete object is built on top of this database, in order to demonstrate its applicability. The algorithm compares the stored values of the principal components with the values computed from a real partial view, searching for the best match of the view's location with respect to the complete object. The paper describes the VPV generation method and shows the experimental results obtained with the pose estimation algorithm. Very low failure rates, even in the existence of occlusions are achieved. These features are demonstrated and shown with several examples performed with real range data.
Important Links:
Go Back
