CAJ | 학술논문

针对变分光流算法的计算精度与鲁棒性问题，提出一种基于图像局部结构的区域匹配变分光流算法。光流估计能量泛函的数据项采用图像结构守恒与灰度守恒相结合，并引入规则化非平方惩罚函数，保证了光流估计的精度与鲁棒性；平滑项采用随图像局部结构自适应变化的扩散策略结合区域匹配约束函数能够有效地保护运动物体或场景的边缘轮廓信息；在光流计算过程中引入金字塔分层细化策略克服图像序列中大位移运动引起的像素点漂移现象，并采用数学方法证明光流估计模型的鲁棒性和收敛性。多组实验表明，本文方法在图像中存在剧烈光照变化、非刚性物体复杂运动以及多目标大位移运动等情况下具有较高的计算精度、较好的鲁棒性。
침대변분광류산법적계산정도여로봉성문제，제출일충기우도상국부결구적구역필배변분광류산법。광류고계능량범함적수거항채용도상결구수항여회도수항상결합，병인입규칙화비평방징벌함수，보증료광류고계적정도여로봉성；평활항채용수도상국부결구자괄응변화적확산책략결합구역필배약속함수능구유효지보호운동물체혹장경적변연륜곽신식；재광류계산과정중인입금자탑분층세화책략극복도상서렬중대위이운동인기적상소점표이현상，병채용수학방법증명광류고계모형적로봉성화수렴성。다조실험표명，본문방법재도상중존재극렬광조변화、비강성물체복잡운동이급다목표대위이운동등정황하구유교고적계산정도、교호적로봉성。
A region matching variational optical flow algorithm based on image local structure according to the computing accuracy and robustness of the variational optical flow method is proposed.Firstly,the data term of the optical flow estimation ener-gy function combines the constancy of the image structure and constancy of the grey value,which introduces a penalty function to ensure the good accuracy and robustness of the optical flow computing.Secondly,the smoothing term uses an adaptive nonlinear dif-fusion function which related to the image structure to preserve the boundaries of the motion.Finally,the region matching strategy and the coarse-to-fine method are used to compute the large displacement optical flow,and then the robustness and convergence are proved by mathematical method.Several experiments prove that the proposed algorithm has the better precisionand robustness even there are terrible illumination changes,complex motion of non-rigid object and large displacement of multiple objects.