Tri-level Unified Framework for Human Gait Analysis

Abstract

There are several applications that can be related to multimedia content analysis. Considering video as one of the prominent forms of multimedia content, this paper presents analysis of human walking motion (gait) found in video sequences by using promising strategy of integrating techniques from data fusion and computer vision. To provide solutions to the challenges in human gait analysis a unified framework is proposed comprising of three different levels: data level, feature descriptor level and decision level. The three levels perform specific tasks assigned to them. At the data level, features are extracted from input video sequences for minimal representation. At the feature descriptor level, features from minimal representation are rearranged to build a feature descriptor and finally at decision level meaningful interpretations are performed. For analysing human walking motion found in video sequences, initially, moving silhouettes are extracted using background subtraction for minimal representation at the data level. The extracted silhouettes are then represented in a common representation in a spatial form followed by correlation analysis and a feature descriptor is developed with minimum interest points at the feature descriptor level. Finally, interpretation of normal gait poses and transition poses are made at the decision level.

Keywords:Multimedia content; Data Fusion; Unified Framework; Background Subtraction;Correlation; Feature Descriptor; interpretation of Gaits.