An efficient algorithm for camera pose estimation in environment of Dunhuang Mogao Grottoes

作者：Luo Weiqun^*; Feng Wei; Jiang Guiyuan; Li Lianmeng; Sun Jizhou

来源：Journal of Computational Information Systems, 2014, 10(2): 837-847.

DOI：10.12733/jcis9642

• Affine transformations
• Autonomous navigation
• Camera pose estimation
• Dunhuang Mogao Grottoes
• Epipolar geometry
• Feature matching
• Fundamental matrix
• Site inspections

The problem of camera pose estimation arises in many areas of computer vision, including object recognition, site inspection and updating, object tracking and autonomous navigation. This paper develops an efficient algorithm to estimate camera pose using two partial overlapped images. In the proposed algorithm, SIFT features are extracted from two overlapped images. Then corresponding features from the two images are matched to produce an initial match set comprises matched features pairs. We develop an novel approach based on affine transformation to exclude mismatch feature pairs from the match set. The match set refining procedure significantly improves the accuracy of subsequent estimation of camera pose. For any combination of eight feature-pairs from the match set, a fundamental matrix can be calculated, we design an efficient approach to calculate the fundamental matrix of minimum residual for camera pose estimation. In addition, we build a virtual system to simulate the camera imaging process in Dunhuang Mogao Grottoes. The system comprises virtual cameras and virtual grottos, and is implemented using OPENGL. Experimental results validates the effectiveness of the proposed algorithm.

• 出版日期2014
• 单位天津大学