单目图像序列中基于正则化的局部三维非刚体运动估计  

Local 3D Non-Rigid Motion Estimation from Monocular Image Sequence Based on Regularization

作　　者：汪亚明[1] 黄文清[1] 李伟[1] 汪元美[2] 

机构地区：[1]浙江工程学院计算机视觉与模式识别研究中心,浙江杭州310033 [2]浙江大学CAD&CG国家重点实验室,浙江杭州310027

出　　处：《电路与系统学报》2004年第2期142-145,共4页Journal of Circuits and Systems

基　　金：国家自然科学基金资助项目(60103016);浙江省自然科学基金青年人才培养项目(RC02064);浙江省自然科学基金资助项目(601019)

摘　　要：通过局部的三维非刚体运动估计进而达到全局估计结果是三维非刚体运动估计中的重要方法。本文提出了在单目图像序列中利用正则化的手段解决局部的三维非刚体运动估计。首先,在帧间特征点匹配已确立的前提下,利用仿射运动模型并结合中心投影方式提出了运动估计的最小二乘模型;然后针对三维运动估计的不适定性提出了正则化的运动估计方法,以正则化的形式融入运动的先验知识,使运动估计的结果更具鲁棒性;最后利用Levenberg-Marquart方法实现运动参数的求解。仿真图像序列的实验反映了本文方法的有效性。Local 3D non-rigid motion estimation is the critical approach for global motion estimation. For this reason, an approach is proposed for 3D local non-rigid motion estimation from monocular image sequence using regularization. First, with the established correspondences of feature points between consecutive frames, the least squares estimation model is proposed based on affine motion model and perspective imaging model. Then, in order to overcome the ill-posed 3D motion estimation problem, a method using regularization is proposed. By incorporating the prior motion knowledge into the estimation process, the more robust result can be achieved. Finally, all motion parameters are estimated by using the method of Levernberg-Marquart. Experimental results from synthetic image sequence demonstrate the feasibility of the proposed approach.

关 键 词：三维非刚体运动估计 单目图像序列 仿射模型 正则化 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]