基于粒子滤波的目标主动轮廓跟踪算法

A Novel Particle Filter Based Object Active Contour Tracking Method

  • 摘要: 传统的粒子滤波方法采用若干维参数定义的简单几何图形给出跟踪结果,不能精确表示现实中具有复杂形状的目标物体。针对这一问题,该文提出基于粒子滤波的主动轮廓算法,用于计算复杂形状目标的轮廓跟踪任务。在目标状态后验分布的模拟样本基础上引入主动轮廓模型,并使用带权粒子定义其能量函数,使得模型的轮廓线向具有重要权重粒子的所在区域演化,并最终收敛到具有最大目标似然的图像区域,从而实现对目标物体的全局运动及局部形态演化的同时估计。精确的目标区域提高了目标模型的更新精度,避免了跟踪中漂移现象的发生。最后,结合真实机场监控验证了该方法在实际复杂场景下的有效性及鲁棒性。

     

    Abstract: Conventional particle filters use simple geometric shapes with finite dimensional parameters to give the tracking results, therefore cannot precisely present the real-world objects with complex shapes. Aiming at this problem, this paper presents a novel particle filter based active contour algorithm for object contour tracking task. The active contour is introduced on the base of the samples simulating the target state posterior distribution; By including the weighted particles into the energy function, the contours evolve towards the region with particles with important weights and eventually converge to image region with maximum likelihood of the target. The proposed algorithm can improve the updating accuracy of the target model and avoid the tracking drift. Finally, a real-world airport surveillance application is presented to show the effectiveness and robustness of the proposed method in complex scenarios.

     

/

返回文章
返回