LACERDA, Y. A.; http://lattes.cnpq.br/5890923112747925; LACERDA, Yuri Almeida.
Resumo:
The knowledge discovery from huge photo repositories has been a very active area of research in the last years. This is due to three facts: the incorporation of digital cameras and geolocation sensors in mobile devices; the advances in Internet connectivity; and the evolution of social networks. The photos stored on those repositories have contextual metadata. Those metadata could be used for many applications of knowledge discovering, such as: Point of Interest (POI) detection; generating of tourist guides; and automatic photo organization. Most approaches for POI detection assume that geographic areas with high density of photos indicate the existence of a point of interest in that area. However, in many cases, the POIs are located in a certain distance of that position in direction where camera was aiming, and not in the exact point of photo shooting. Most of related work do not consider the use of orientation in the process of POI detection. In this way, we propose a set of algorithms and techniques for POI discovery in touristic cities using geotagged and oriented photos collection exploring the geographic orientation in different ways. This research has proven the importance of the usage of orientation in the new algorithms for POI detection. In the experiments with collections related to big cities, the algorithms considering orientation, in several scenarios, have beaten those that do not consider. Also, new metrics of evaluation have been proposed and a new framework to assist all the tasks for knowledge discovery based on huge photo collections.