Multi-Person Tracking using Overlapping Silhouette Templates for a Mobile Robot

我们开发了一种跟踪方法，该方法使用应用于密集深度图像的人的形状的深度模板。一个准确的，稳定的跟踪是通过使用重叠的剪影模板，考虑如何在图像中的人重叠。通过使用基于核密度估计的近似似然图，减少了图像比较的次数。我们在有许多普通人存在的真实的环境中进行了特定人跟随的实验，并对结果进行了分析。

We developed a tracking method which uses depth templates of person shape applied to a dense depth image. An accurate, stable tracking is realized by using overlapping silhouette templates which consider how persons overlap in the image. The number of image comparison is decreased by using an approximated likelihood map based on kernel density estimation. We performed experiments of specific person following in real environment where many ordinary people exist, and analyzed the results.