Contributions to object detection and human action recognition

Resumen

The amount of available images and videos in our everyday life has grown very quickly in the last few years. Mainly due to the proliferation of cheap image and video capture devices (photo cameras, webcams or cell phones), and the spread of the Internet accessibility. Sites for photo sharing like Picasa or Flickr; social networks like Facebook or MySpace or video sharing sites like YouTube or Metacafe, offer a huge amount of visual data ready to be downloaded in our computers or mobile phones. Currently, most of the searches, performed in online sites and on personal computers, are based on the text associated to the files. In general, the textual information is usually poor compared to the rich information provided by the visual content. Therefore, it is necessary efficient ways of searching photos and/or videos in collections, making use of the visual content encoded in them. This thesis focuses in the problems of automatic object detection and categorization in still images, and the recognition of human actions on video sequences. We address these tasks by using appearance based models.