提出了一种由单轴运动下的未标定运动图像序列进行三维重构的新方法。利用不变量计算出第一幅图像对应的摄像机矩阵,根据转台的旋转角度计算出其余图像对应的摄像机矩阵,再使用SIFT算法查找相邻两幅图像可见的匹配点,并利用线性三角形法对图像进行重构,最后用捆集调整优化重构的结果。真实图像序列的重构实验验证了算法的正确性和有效性。
The paper presents a new approach for recovering 3D geometry from uncalibrated image sequences under single axis motion.It computes camera matrix corresponding to the first image by use of invariant,acquires camera matrixes of the other images according to rotation angles of turntable.Then it searches visible match points between adjacent images by SIFT algorithm,recovers 3D structure according to linear triangulation.Finally,it makes use of bundle adjustment to optimize the reconstruction results.The experiments on real image sequences demonstrate the correctness and effectiveness of the method.