H.264 encoder based 3D model acquisition and compression for aerial video

In this paper, an automatic 3D model extraction from aerial images is proposed. The system is based on the H.264 video encoder that is used for 3D information extraction and compression. In order to obtain a disparity map from the sequence of aerial images, motion vectors received from the H.264 encoder motion estimation algorithm are used. The main drawback using motion estimation for disparity generation is computational noise. A new statistical approach is proposed to estimate the disparity of building surfaces from a noisy motion estimation map. A method for disparity compression for further transmission is also suggested. The proposed algorithm can be integrated in the H.264 encoder for fully automatic 3D urban scene extraction and compression. Tests are performed on real video sequences taken from several sites from various airborne camera heights.