基于水平集的多运动目标时空分割与跟踪

针对背景运动时的运动目标分割问题,提出了一种对视频序列中的多个运动目标进行分割和跟踪的新方法。该方法着眼于运动的且较为复杂的背景,首先利用光流约束方程和背景运动模型建立一个基于时空域的能量函数,然后用该函数进行背景运动速度的估算和运动目标的分割和跟踪。而时空域中的运动目标的最佳分割,乃是通过使该能量函数最小化来驱动时空曲面演化实现。时空曲面的演化采用了水平集PDEs(Partial Differential Equations)方法。实验中,用实际的图像序列验证了该算法及其数值实现。实验表明,该方法能够同时进行背景运动速度的估算、运动目标的分割和跟踪。

A Spatiotemporal Multiple Moving Objects Segmentation and Tracking with Level Set

Aimed at the moving objects segmentation problem with a moving background,a method of segmentation and tracking of moving objects was proposed. Firstly,this method built up a spatio-temporal energy function based on the optical flow and the motion model of background.Then,the segmentation,tracking and estimation of the motion of background were processed using the energy function.Minimization of the energy function led to the optimal segmentation of moving objects in spatio-temporal domain by curve evolution.The level sets PDEs(Partial Differential Equations) approach was used for the evolvement of the curves in spatio-temporal domain.This algorithm and its numerical implementation were verified on real image sequences.Experimental results show that the method allows the segmentation and tracking of multiple motion with estimating simultaneously the motion of background.