关键词: 未标定图像, 双向匹配, 三维重建, 图像序列, 四元数

Abstract: This paper proposed an algorithm for 3D reconstruction based on uncalibrated images. First, this algorithm detected the feature points by using Harris corners, and presented an improved bidirectional matching to match these feature points. Under the condition of the known camera parameters, the authors carried out 3D reconstruction of two images. Then the quaternion algorithm was adopted to transform coordinate. The results coming from different subsets were brought into a common coordinate frame, achieving 3D reconstruction of multiple images. Finally, bundle adjustment was used to optimize the results. The experimental results show that object structure can be well reconstructed by using this algorithm.

Key words: uncalibrated image, bidirectional matching, 3D reconstruction, image sequence, quaternion