基于Hausdorff距离的人体步态识别

现有的步态识别方法对行人轮廓匹配的鲁棒性差，导致识别率不高．文章提出了基于Hausdorff距离的行人步态自动识别方法．首先提取了行人二值轮廓序列；然后采用轮廓参考点分布直方图间的距离、参考点集之间Hausdorff距离度量轮廓形状问的匹配度；继而通过步态的周期性分析选取关键姿态，计算出的关键姿态轮廓集间Hausdorff离结合窗口搜索策略实现了步态的分类和识别．分别在小型CASIA室外步态数据库和大型Soton室内库上进行了实验，提出算法的正确识别率分别可达到91．25％和88．16％．与相关文献的比较分析表明算法是有效的．

Automatic Gait Recognition Based on Hausdorff Distance

A new method for automatic gait recognition by computing the Hausdorff distance between two sequences of walking figures is proposed.For each image sequence, the binary silhouettes of a walking figure were first obtained by an improved background subtraction algorithm.The silhouette was transformed into a distribution histogram-based feature vector sets consisting of reference points sampled from the contour.Hausdorff distance between test and reference key pose was then assessed.Combined with window shift technology, Hausdorff distance between test and reference sequences was executed to produce similarity score.Subject classification and recognition were finally performed by nearest neighbor matching decision.Experimental results demonstrate that the proposed approach has an encouraging recognition performance with respect to other approaches on these databases.