基于组合模板的人体头部精确定位算法

A Novel Head-location Algorithm Based on Combined Mask

对人体头部的准确定位是当今计算机视觉研究中一个非常重要的领域。在基于视觉特征的人体头部提取算法的基础上,提出了一种改进的算法——"粗—精"双步头部提取算法。该算法首先利用人体头部与躯干相对关系初步确定人体头部位置及大小,然后通过水平和垂直方向的两个精确定位模板对粗定位结果进行调整,最终达到精确定位的目的。详细阐述了"粗—精"双步头部提取算法的基本原理和实现方法,并针对行走中的人体,给出了改进前后两种算法实验效果。实验结果表明,改进后的头部定位算法可以在复杂情况下更精确地定位头部,鲁棒性强。

Human head location plays an important role in applications such us video surveillance, human activities analysis, perceptual user interface and so on. In this paper, A novel algorithm, called ＂coarse-accurate＂ two-step head location, based on view-based human head location algorithm, is proposed. The proposed approach first locates the head＇ s coarse position depending on the relative relationship of the human head and torso, then adjusts the coarse position to the accurate position by adopting two novel masks. In the last part of this paper, the results of improved and unimproved algorithms are compared. The experimental results show that the proposed algorithm is more robust in complex conditions and can detect the head more accurately.