3D Modelling of Cultural Heritage Objects from Photos Posted over the Twitter

The tremendous increase of tweets messages, also posting visual information and photos of cultural heritage (CH) sites, opens new ways for filtering and 3D modelling this type of content. This is very important in documenting CH objects. This paper firstly applies novel event detection algorithms from tweet signals exploiting retweeting and followers properties. The temporal signal is then decomposed using the Discrete Wavelet Transform enabling localization both in time and frequency domain. Multi-assignment graph partitioning takes place to identify the events. Upon this textual constraint, a novel content-based filtering scheme is introduced to cluster the images into appropriate groups which in the sequel trigger 3D reconstruction. Experimental results on real-life data reveal the superiority of this method than conventional approaches.