Construction d’une mosaïque à partir d’une séquence vidéo

Abstract

In this thesis, we study the stages of the construction of an image mosaic, an important problem in the field of computer vision. First, we evaluate the existing methods, which deal with all the stages of construction of a mosaic. Next, we propose a new approach for the mapping by using the technique of comparison by triangles. The construction of a mosaic is done in four stages: the detection of the points of interest, the correspondence mapping, the calibration of the camera and the projection of the images on a cylindrical projection plane. To achieve the first stage which consists in extracting the points of interest, we adopted the algorithm of Harris. The choice of a threshold value to select the points of interest is a difficult problem. To improve the distribution of the points extracted in the image, we propose a technique of “adaptive thresholding”. In the second stage, we propose our approach of mapping that uses the points of interest obtained in the first stage to identify the pairs of the mapping. To achieve this goal, we work out a method called “method of comparison by triangles” which adds a “geometrical” dimension to the choice of the pairs of corresponding points. Finally we combine several images processed through the three preceding stages to obtain the mosaic by using cylindrical projection.

Documents

PDF

---