Abstract |
The gradual development of video as mean of information transmition has created new needs for storing, transmitting and processing of its content. The lack of representation, manipulation and editing tools of video sequence doesn't allow the full exploitation of its content. This can be achieved by the construction and use of an image or a number of images that represent the static background of scene. These images are the panoramic views of the scene. They are also known as mosaics. In our work, we propose a method of constructing these images. The basic steps of the construction of panoramic views are the frame alignment, which is based on the camera motion estimation, and the frame integration of the sequences. The camera motion is described by the well known parametric models: projective, affine, ``zooming and transmitting" models. The estimation of motion parameters is based on the use of robust methods such as the M-estimator algorithm or on affine moment invariants. The implementation of these methods requires the determination of the optical flow field that can be resulted by the block matching or corner correspondence. Variation of steepest descent and relaxation methods were used for the correction of motion parameters. The frame integration can be achieved by frame overlapping or by using mean and median value filters. The problem of false alignments of the construction of mosaics was solved by frame topology. Experimental results are presented in order to indicate the efficiency of the above methods.
|