An Automatic Indexing Method of Video Contents using 3-Dimentional Geographic Data - Automatic Extraction of Buildings for Indexing

摘要

This paper proposes an automatic extraction method of buildings shot by a video camera for indexing purpose so that users can retrieve video data by content-based specification. Video are taken by a camera with an wearable computer attached by a GPS and a gyro-sensor. A set of shot buildings are automatically extracted using a 3-dimentional geographic data stored in the wearable computer. Video stream are divided into "unit" where the set of shot building does not change. By this means, the system makes it possible for user to retrieve units whose contents are, for example, "Ginza Mitsukoshi Department Store in the evening darkness" from video database.