Improvement of Accuracy of 3D Structure Reconstruction from Interlaced Video Sequence

Abstract

This paper investigates a 3D reconstruction of the building by detecting the vanishing points from the interlaced video sequence. It is theoretically possible to reconstruct a 3D model of the building from a single picture. However, the accuracy of reconstruction is not enough with a single picture because of the deterioration of the picture, such as noise or dim etc. Increasing the number of pictures, accuracy of the reconstructed model will become higher. It is also possible to apply this method to video sequences. There are deteriorations caused by interlace scanning in addition to noise in each video image. Our preliminary results show that from information of the many pictures, this method can reconstruct precise 3D structure. The addition average values that are obtained from the 50 pictures or more have been settled within 2% of the actual value.