Learning Human Identity Using View-Invariant Multi-view Movement Representation

Abstract

In this paper a novel view-invariant human identification method is presented. A multi-camera setup is used to capture the human body from different observation angles. Binary body masks from all the cameras are concatenated to produce the so-called multi-view binary masks. These masks are rescaled and vectorized to create feature vectors in the input space. A view-invariant human body representation is obtained by exploiting the circular shift invariance property of the Discrete Fourier Transform (DFT). Fuzzy vector quantization (FVQ) is performed to associate human body representation with movement representations and linear discriminant analysis (LDA) is used to map movements in a low dimensionality discriminant feature space. Two human identification schemes, a movement-specific and a movement-independent one, are evaluated. Experimental results show that the method can achieve very satisfactory identification rates. Furthermore, the use of more than one movement types increases the identification rates.