Research On The Key Technology Of Sequential Three-Dimensional Reconstruction Independent Of Priori Knowledge

Three-dimensional reconstruction based on image is a key problem in Computer Vision and Photogrammetry.Through the unremitting of a large number of researchers,the mathematical basis of three-dimensional reconstruction has tend to be prefect,and the technical process of three-dimensional reconstruction has been finalized.At present,it is the major research direction of structure from motion that how to quick and stable reconstruction high-accuracy three-dimensional model without priori knowledge and control point.It has been the emphasis and difficulty of three-dimensional reconstruction based on image,because of the reconstruction process without priori knowledge and uncertainty.In this paper conducts a thorough study into the sequential three-dimensional reconstruction strategy,the main contents of this paper include:1.We realize the sequential three-dimensional reconstruction process base on the previous research.Then,we have a lot of experiment with each process of threedimensional reconstruction,and analysis of experiment data and find out the problem in three-dimensional reconstruction process.Finally,analysis the cause of the problem and try to improve it.The experimental results show the feasibility of this methods.2.The limitation that still exists in the existing three-dimensional reconstruction algorithm,such as some images can not add to reconstruction and the global bound adjustment often fails in the pair which have weakly connected.Based on this,we proposes a new three-dimensional reconstruction strategy which adopts a new image selection strategy and adaptive threshold global bound adjustment.The image selection strategy first has a local adjustment to the new join points,then select the qualified connection points that their re-projection error is less than the limit;finally,select the image to be added,according to the number of connection points after adjustment.The threshold estimates by the local bound adjustment report,in order to ensure the global bound adjustment success.Experimental results proof that our method can improve the stable of the three-dimensional reconstruction and reduce the requirement of two view geometry based on experience.The algorithm has good performance of the pair which have weakly connected.