采用显著性分割与目标检测的形变目标跟踪方法
A Deformed Object Tracking Method Utilizing Saliency Segmentation and Target Detection
-
摘要: 在跟踪形变目标时,为了克服采用颜色特征分割和学习目标模型而产生跟踪漂移的问题,提出一种基于显著性分割与目标检测的跟踪方法.该方法采用图流形排序方法对超像素进行显著性分割来获得显著、高质量的目标像素,并使用这些像素的颜色和梯度量化值及其与目标中心的相对位置关系来表示目标模型,通过目标模型检测目标中心位置实现跟踪;以全局颜色特征作为漂移约束条件,对由目标持续形变而产生的中心位置漂移进行修正;在跟踪过程中,将跟踪结果作为正样本对显著分割和全局颜色特征进行约束,依据显著分割结果更新目标模型.实验结果表明,文中方法有效地提高了形变目标跟踪的精度和稳定性.Abstract: This paper addresses the tracking drift problem in deformable target tracking, which is caused by the target segmentation and modeling with color features. To this end, we propose a novel tracking method based on saliency segmentation and target detection. We utilize graph-based manifold ranking to obtain salient and high-quality target pixels, and quantify these pixels by color and gradient. Then, the quantized values of target pixels and the relative position to the center of the target are used to represent the target model, which is used for tracking target by detecting the target center. A drifted constraint based on the global color features is used to correct the central position drift caused by the continuously deforming the target. Moreover, the tracking results are adopted as the constraints of saliency segmentation and global color features, and the target model is updated according to the segmentation results. The experimental results demonstrate that the proposed method is able to improve the precision and stability for deformable target tracking.