Análisis de la marcha de reconocimiento basado en el jacobi

Gait recognition analysis based on the Jacobi-Fourier moment histories
C. Camacho-Bello, C. Toxqui-Quitl∗ , A. Padilla-Vivanco and C. Santiago-Tepantl´n a
´ Laboratorio de Optica y Visi´n por Computadora. Universidad Polit´cnica de o e Tulancingo. Ingenier´ 100, Huapalcalco 43629, Tulancingo, Hidalgo, M´xico. ıas e
∗

ctoxqui@upt.edu.mx

Abstract A new method to compute the temporalcorrelation in terms of a moment function representation of lower body images for gait recognition is presented. A Jacobi-Fourier moment history (JFMH) is obtained from N -frames to describe the history shape of individuals. The kernel of the moments is composed by the Fourier factor exp(jlθ) of order l and the Generic radial Jacobi function Jn (α, β, r) of order n. Value variations for α and β generatean infinite number of orthogonal polynomial families. With this in mind a performance analysis of different circular moments is done. The automatic phasing of several moment histories using genetic algorithms is implemented for video temporal calibration. A discriminative method of moments by means some selective n − l maps is proposed. Descriptors obtained from a selected JFMHs in phase have asmall intraclass variance and large interclass separation. By using the combined metrics of correlation and minimum distance classifier M -classes, our method fitting only the selected features in order to obtain the highest Correct Classification Rate (CCR) for all sets tested. The Jacobi-Fourier descriptors based on polynomials Jn (10, 10, r) achieve a full recognition percentage with the lowestnumber of descriptors compared with others as Zernike. The databases used here are MoBo, CASIA A, and our home database. Keywords: Gait biometrics, Jacobi - Fourier moments, Genetic algorithms, invariance features.

Preprint submitted to Image and Vision Computing

May 5, 2011

1. Introduction From early ninth decade, several algorithms to extract human gait features from image sequences havebeen studied for individual recognition [1], [2], [3]. They have looked for a specific pattern such as style of walk, or pathology [4]. Typically, a model-based approach for human gait is required to estimate certain parameters such as gait frequency, phase, and center of mass coordinates [5], [6]. However, this procedure requires a large number of operations for a single image. Although this mightbe a more accurate model is also computationally expensive. Other possibilities can use holistic measures as Zernike moments [7]. The main advantage is that, an holistic method can be computationally cheaper than a model-based approach. In this paper a novel holistic measure is proposed. It is based on the Generic Jacobi-Fourier Moment History (JFMH). Bathia and Wolf [8] demonstrated that thereexist an infinity number of Jacobi Fourier polynomial sets. Each set is obtained by the combination of the parameters α and β [10]. Thus, a performance analysis is done using some orthogonal polynomial families. Also JFMH can be adapted by a geometrical moment process to generate invariants to translation, illumination, small rotations, and scale of walking people. Advantages of using the method oforthogonal moments is that they can obtain different phases of gait by means moment histories from image sequences. Here, a general problem is to find a initial phase of gait in the sequence. It is not unknown that, there exist research about correlation for measuring frame similarities between subsequences and a target sequence in gait and action recognition areas [11], [12]. Nevertheless, thecomputational cost of this three dimensional correlation in the space and time domain is high, especially when the size of images is large. Murase [13] proposes to reduce the computational cost with a eigenspace representation of the image and then, the 2D correlation of eigenvectors sequences is done. In this paper, we propose maximize the 1-D correlation of several moment histories using genetic...