针对摄像机运动的情况,提出多目标分割和跟踪的新方法.利用主动轮廓模型,将运动估计和运动分割融合在同一基于时空域的能量泛函中.为了克服传统的活动轮廓模型和水平集方法存在的局部最小值问题,对时空分割模型进行凸优化,避免了初始化轮廓对分割结果的影响,保证了能量函数对分割的全局最优性.提出相应的快速原始-对偶算法,提高了计算效率.实验表明,该方法能够有效地实现运动场景下的时空域运动分割与跟踪.
A new method for multi-target segmentation and tracking with a moving viewing system was proposed.The active contour model was used for integrating motion estimation and segmentation into a variational framework throughout the spatiotemporal domain.A global convex minimization method was applied to the spatiotemporal tracking model,overcoming the limitation of the level set method that is sensitive to the initial condition.The primal-dual algorithm was designed for the proposed model in order to improve the computation efficiency.Experimental results illustrated the validity and improvements provided by the proposed spatiotemporal tracking model.