A scalable gait acquisition and recognition system with angle-enhanced models

Person recognition through gait is a highly promising biometric technique, offering substantial advantages over traditional methods. Despite its potential, gait recognition from images can be challenged by factors such as variations in viewing angles, personal accessories, or clothing, which may alter specific gait characteristics. A novel and innovative gait identification system with two main components was developed to address these challenges. The first component focuses on the acquisition of gait sequences, using algorithms for detection, tracking, and gait analysis from images. The gait analysis algorithm facilitates the extraction of high-quality image sequences while determining the subject's movement angle relative to the imaging camera. This is essential for ensuring precise and consistent data for the identification process. The second component is the person identification algorithm, which employs model-free approaches. This component includes various approaches to integrate the angle information into four well-established models: GaitPart, GaitSet, GaitGL, and GaitBase built using the CASIA-B dataset. The results demonstrated that angle information can refine feature extraction when properly integrated into the model, achieving state-of-the-art results across the four models. The GaitPart, GaitSet, and GaitGL models preferred late-stage angle integration, whereas GaitBase performed better with early-stage integration due to its strong backbone. In the final phase of this study, additional tests were conducted using the modified GaitBase model with angle information on the CASIAE dataset. These tests confirmed the model's effectiveness and enabled a detailed analysis of the threshold that differentiates gait sequences from the same person and those from different individuals. This threshold enhances the system's scalability by enabling it to determine whether a person has been previously observed. Thus, this study developed an innovative and theoretically scalable system adaptable to a growing number of users and locations, with potential applications in access control, security monitoring, and attendance management.