全局光流场估计技术及展望

Technology and Prospect of Global Optical Flow

光流法是运动分析最为重要的技术之一,其能够从相邻帧图像中恢复出目标物体以及背景的运动信息,从而实现目标检测、运动跟踪以及特征识别等.文中从全局光流计算的基本原理出发,详细分析了非均匀光照校正、能量函数构建和优化方案设计等全局光流场估计技术的关键步骤;同时深入探讨了全局光流场估计技术存在的大位移计算、遮挡检测、弱纹理运动估计等难题.最后对全局光流场估计技术所面临的挑战进行了分析,并展望了其未来的发展方向.

Optical flow is one of the most important technologies for motion analysis, which can reconstruct the motion of objects from consecutive image frames and hence can be utilized for object detection, motion tracking and feature identification. Based on the principle of global optical flow, the main techniques are reviewed in detail from three sections： non-uniform illumination correction, energy function construction and optimization scheme design. Meanwhile, three main difficulties for global optical flow analysis are studied thoroughly, which include large displacement computation, occlusion detection and weak texture estimation. At last, challenges are summarized and prospects of this technique are made.