摘要
基于随机采样最小冗余子集新概念,本文开发了一个从目标的单视图特征点集合鲁棒精确地复原其三维位姿的新算法。在强噪声高出格点率的恶劣条件下,该算法仍能高精度地复原目标的三维位姿。实验表明,对于由100个点组成的单视图特征点集合而言,当出格点率高达90%并且内点信噪比低达28db时,它仍能以1%的相对误差复原目标特征点的三维坐标。
Based on the new idea of the minimal redundant subset of random sampling, we develop a new algorithm to robustly and accurately determine the 3D pose of an object from its model data and its image data, which consist of the key point set extracted from its single perspective view. The algorithm is able to work well with high accuracy in the case of heavy noise and high outlier rate. The experiments have demonstrated that for image data set consisting of 100 key points, the algorithm is still able to restore the 3D coordinates of the key points with relative error of 1 % , even if the outlier rate is as high as 0. 9 and the SNR of its inliers is as low as 28db.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
2001年第1期38-41,共4页
Pattern Recognition and Artificial Intelligence
