| 摘要: |
| 复杂环境下的多目标视频跟踪是计算机视觉领域的一个难点，有效处理目标间遮挡是
解决多目标跟踪问题的关键。提出了一种融合遮挡分割的多目标跟踪算法，计算每个目标
的光流速度概率直方图，反映其运动统计信息；综合使用外观、运动、颜色信息构造新的像
素距离表达，借助分阶段分类思想及K均值聚类技术进行遮挡分割，得到准确的运动前景
像素；在粒子滤波器跟踪框架下，使用概率外观模型进行多目标跟踪，更好地处理动态遮挡
问题。实验表明，所提算法解决了复杂环境下的多目标跟踪问题。 |
| 关键词: 多目标跟踪 遮挡分割 粒子滤波 概率外观 光流速度直方图 K均值聚类 |
| DOI： |
|
| 基金项目: |
|
| A multi-target tracking algorithm combined with occlusion segmentation |
| LI Jian-yong,XU Lian-yu |
| () |
| Abstract: |
| Multitarget tracking in complex scenes is one of the most complica
ted problems in computer vision.Handling the occlusion between objects is the k
ey issue in multitarget tracking.An occlusion segmentationbased algorith
m
is presented to track multiple people through complex situations which are capt
u
red by static monocular cameras.In the proposed algorithm, the pro
babilistic histogram of each object′s optical flow vector is calculated,then t
his motion
statistic information along with the color and appearance information is used t
o constr
uct a new expression of pixel distance,finally stepwise classification and Km
e
ans clustering method are taken advantages of to accomplish occlusion segmentati
on.Object is handled by a particle filterbased tracking framework,and a prob
a
bilistic appearance model is used to find the best particle.Results show that t
he proposed approach can improve the performance of the original probabilistic a
ppearance model and handle dynamic occlusion better. |
| Key words: multi target tracking occlusion segmentation particle filter probabilisti
c appearance optical flow histogram K means clustering |
