Título: Robust image segementation towards an action recognition algorithm
Autores: Arseneau, Shawn.
Fecha: 2000
Publicador: McGill University - MCGILL
Fuente:
Tipo: Electronic Thesis or Dissertation
Tema: Computer Science.
Descripción: To facilitate proper recognition of a human's action from a video sequence, several key features must first be determined. Initially, the person performing the action must be isolated from the background scene. This information is then used to decipher pertinent action attributes that may include the center of mass, contours, and regions of motion. It is these characteristics that will become the feature elements in the recognition of a person's actions.
This thesis will investigate the various image processing tools available to obtain the aforementioned action attributes. The applicability of filters, background removal techniques, skin-tone matching, and contouring schemes will all be investigated. A thorough comparison with both existing and novel approaches to action recognition is then discussed. Overall, the temporal based algorithm is best suited for an action recognition application as the spatially based approaches rely too heavily on a priori knowledge of the background scene.
Idioma: en