A practical method of 3D reconstruction based on uncalibrated image sequence

Abstract

3D reconstruction is an important technique in the domain of computer vision. The existing systems of 3D reconstruction are often built on specialized hardware such as structured light or laser scanner with high cost. And some systems require intricate calibration procedures using complex calibration objects. These are not practical to reconstruct the 3D model of an arbitrary scene. In this paper, a flexible method of 3D reconstruction based on uncalibrated image sequence is proposed. This system consist a series of processes such as feature points extraction and match, projective reconstruction, self-calibration, dense match, and finally the dense surface of 3D model can be reconstructed. Experimental results prove that this method is practical and has higher precision.