基于三焦点张量的多视图三维重构

3D Reconstruction of Multiple Views Based on Trifocal Tensor

通过二维图像恢复三维物体的几何信息在整形外科、骨科等有重要的应用。本文提出一种基于三焦点张量的多视图三维重建方法。该方法不需要摄像机在拍摄多视图时保持纯平移或者纯旋转运动。摄像机任意拍摄多幅图像,根据三视图之间的几何关系进行射影重建,然后通过在分层重建中引入cheirality约束确定无穷远平面的搜索范围,以完成摄像机内参数的标定,并最终实现度量重建。本文还对重建后的模型进行了鼠标拾取坐标点和键盘响应,以对模型进行各个角度的观察。利用真实物体图像序列以及人脸数据图像进行了实验,实验结果表明该方法具有较好的鲁棒性和准确性。

Reconstruction of 3D structure of an object from 2D views plays an important role in plastic surgery and or- thopedics. This method doesn＇t need camera to do specific movements, such as translation or rotation independent- ly. It only needs a hand-hold camera arbitrarily to take a few pictures, and apply the geometry relationship among the three views to obtain the projective reconstruction of the object. Then, it needs to introduce cheirality constraint in stratified reconstruction to determine the search area of the infinity plane, and finally achieve the camera intrinsic pa- rameters calibration, and complete the metric reconstruction. This model has also been reconstructed with mouse and keyboard response coordinates to observe the model from different angles. Experiments with both pictures of object and face pictures show that the proposed method is very robust and accurate.