基于DRLSE模型的运动目标跟踪

高斯粒子滤波不能处理曲线的拓扑变化,而基于水平集的几何活动轮廓模型能较好地适应拓扑变化,为了跟踪和提取刚体和非刚体运动目标精确的轮廓信息,提出基于距离规则化的水平集演化(DRLSE)模型和高斯粒子滤波(GPF)相结合的运动目标跟踪方法.首先用高斯粒子滤波对目标跟踪得到目标的运动区域,然后把水平集规则项引入到测地线活动轮廓模型中,以外接轮廓的中心为基础进行DRLSE.距离规则化不仅消除了水平集重新初始化的需要,而且避免了因此而导致的数值错误,在水平集演化过程中保持了水平集函数的规则化.最后,将获得的精确轮廓信息反馈到跟踪框架.实验结果表明,该方法适用于刚体和非刚体目标,在实际交通环境中跟踪结果更加精确.

Moving Object Tracking Using DRLSE Model

Gaussian particle filter(GPF) can not process topology changes of curves,while geometric active contour model based on level set can well adapt to the topology changes.In order to track and extract accurate contour information of rigid and non-rigid moving objects,a moving object tracking method combining distance regularized level set evolution(DRLSE) model and GPF is proposed in this paper.Firstly,moving regions of the objects are obtained through object tracking with GPF.Then the level set regularization term is introduced to the geodesic active contour model and DRLSE is performed by taking external contour center as the basis.The distance regularization effect eliminates the need for re-initialization and avoids the corresponding numerical errors.The regularity of the level set function is maintained during the level set evolution.Finally,the obtained accurate contour information is feed back to the tracking framework.Experimental results show that the proposed method is applicable to the rigid and the non-rigid objects and tracking results are accurate in real traffic environments.