Ph.D. Tezi Görüntüleme

      Biometric systems capable of performing reliable, fast and autonomous human identification and identity verification focused scientific attention on biometric feature extraction especially in criminal cases and other events demanding high level of security. The goal of recent studies on face and gait recognition is to foresee and prevent human-based unlawful incidents such as terrorism threatening security. Gait is a biometric feature with differentiating characteristics including remote profile formation by using low resolution video cameras, measurability in dark and complicated environments via thermal and infrared video cameras, and imitation and concealment hardness. Due to its distinguishing attributes, in the near future, human identification based on gait is expected to serve as the core component of biometric systems located at critical security points.

      This work focused on gait recognition based silhouettes. Walking signatures have been created using feature extraction methodologies on gait patterns. The similarities among walking signatures have been estimated using nearest-neighbor and support vector machine learning algorithms. This study contributed to the areas including gait pattern production, automatic parameter estimation for Gaussian kernel, attribute normalization and distance calculation. Furthermore, direct linear discrimination and its kernel approaches have been applied for feature extraction. By means of the application of the methods developed in this work which can be adopted in real-time systems, a performance increase has been observed in widely used CMU, SOTON, USF and CASIA databases.