Research Of Some Gait Feature Representations

Gait, referring to the way people walk, is shown by psychological experiment and anatomy theory to be distinct between different persons, and thus can be used for human identification purposes. The advantages of using Gait include, but is not limited to, gait information is obtainable from far distance, is easy to get, is hard to conceal or forge, and requires no physical contact with the subject. All the advantages make gait a perfect choice in many security surveillance scenarios, and therefore have brought gait much attention in recent years.The research topics of gait recognition includes gait data acquisition, target seg-mentation, gait feature extraction and gait classification, in which gait feature extrac-tion is the most challenging one, and is the focus of this thesis. This thesis mainly contains the following content:(1) introduces the concept of Dimension Reduction, a key issue in pattern recog-nition research, and discuses about two important and widely used techniques:P-CA(Principal Component Analysis) and LDA(Linear Discriminant Analysis).(2) proposes an improved contour-centroid distance frequency analysis based gait feature representation, which in two perspectives makes use of gait's periodicity. Ex-periments on multiple datasets are performed and good results are obtained.(3) proposes an improved contour-projection signal statical analysis based gait feature representation, which does not require sampling different periods so that they have the same number of silhouettes, and thus avoids losing gait information. Experi-ments on multiple datasets are performed, and good results are obtained.(4) proposes a novel texture based gait feature representation. This representa-tion is robust, has a minimum reliance on silhouettes'quality, and specifically can be used for dealing with low quality silhouettes. Experiments on multiple datasets are performed and good results are obtained.