Gait Analysis in Unconstrained Environments

Gait can be defined as the individuals? manner of walking. Its analysis can provide significant information about their identity and health, opening a wide range of possibilities in the field of biometric recognition and medical diagnosis. In the field of biometric, the use of gait to perform recognition can provide advantages, such as acquisition from a distance and without the cooperation of the individual being observed. In the field of medicine, gait analysis can be used to detect or assess the development of different gait related pathologies. It can also be used to assess neurological or systemic disorders as their effects are reflected in the individuals? gait. This Thesis focuses on performing gait analysis in unconstrained environments, using a single 2D camera. This can be a challenging task due to the lack of depth information and self-occlusions in a 2D video sequence. The Thesis explores the use of gait, to perform biometric recognition and pathology detection and classification by reviewing the state-of-the-art and presenting novel taxonomies to organise the systems. In the field of biometrics, the work done in this Thesis improves the performance of the recognition systems by proposing two novel gait representations. It also addresses the problems faced by recognition systems in unconstrained environments, such as change in the viewpoint of the camera and change in the appearance of the individuals being observed, presenting three novel systems to detect the viewpoint of the camera and a system to tackle appearance change. Finally, the Thesis explores the possibility of obtaining gait features from the shadow cast by the individuals, presenting two systems to rectify the distortion and deformation in the shadow silhouettes and a system to detect if the shadow is usable. It also presents two datasets to evaluate these systems. In the field of medicine, this Thesis presents a novel system to obtain biomechanical features, from a video sequence captured with a 2D camera, with a high level of accuracy, while also being robust to viewpoint change. To evaluate the system the Thesis presents a dataset containing sequences acquired from a 2D camera and the ?gold standard? motion capture system. The Thesis also explores the ability of gait to classify different gait related pathologies. It presents two novel systems that perform classification of gait across different gait related pathologies using biomechanical features and deep convolutional neural networks. A comprehensive evaluation of the proposed systems and comparison with the state-of-the-art highlight the advantages of the proposed systems for biometric recognition and pathology classification.