Key Frame Selection Algorithms for Automatic Generation of Panoramic Images from Crowdsourced Geo-tagged Videos

摘要

Currently, an increasing number of user-generated videos (UGVs) are being collected - a trend that is driven by the ubiquitous availability of smartphones. Additionally, it has become easy to continuously acquire and fuse various sensor data (e.g., geospatial metadata) together with video to create sensor-rich mobile videos. As a result, large repositories of media contents can be automatically geo-tagged at the fine granularity of frames during video recording. Thus, UGVs have great potential to be utilized in various geographic information system (GIS) applications, for example, as source media to automatically generate panoramic images. However, large amounts of crowdsourced media data are currently underutilized because it is very challenging to manage, browse and explore UGVs. We propose and demonstrate the use of geo-tagged, crowdsourced mobile videos by automatically generating panoramic images from UGVs for web-based geographic information systems. The proposed algorithms leverage data fusion, crowdsourcing and recent advances in media processing to create large scale panoramic environments very quickly, and possibly even on-demand. Our experimental results demonstrate that by using geospatial metadata the proposed algorithms save a significant amount of time in generating panoramas while not sacrificing image quality.