Double-Bounding R-tree: A Novel Approach to Index Digital Camera’s
Viewable Scene
He Ma, Yue Zhao, Guohui Wei, Xiaoyu Cui
Sino-Dutch Biomedical and Information Engineering School, Northeastern University, Shenyang,
110169, China
email: {mahe, cuixy, zhaoyue}@bmie.neu.edu.cn
Keywords: Video Indexing, Meta-data Management, Query Processing
Abstract. Geographic Information System (GIS) applications now increasingly make use of
geo-located multimedia data such as images and videos. Furthermore, the widespread use of
smartphones (and increasingly tablets) and the rapid improvement of hardware has enabled the
acquisition of high-definition user-generated videos that are annotated with geo-properties. The
sensor meta-data (e.g., GPS and digital compass values), which associate a continuous stream of
location and viewing direction information with the collected videos, are considerably smaller in
size than the visual content and are helpful in effectively and efficiently manage and search through
large repositories of videos. In this paper, we proposed the Double-Bounding R-tree (DBR-tree),
which utilizes the inner-bounding rectangle and outer bounding rectangle to manage the camera’s
field of view. Experimental results show that this structure can prune the unnecessary overlap
calculation and improve the spatial query performance.
Introduction
Recent developments in video recording technology have enabled user-generated video (UGV)
production on a daily basis. For example, smartphones (and increasingly tablets), which are carried
by users all the time, have become extremely popular for capturing and sharing online videos due to
their convenience, enhanced quality of images and wireless connectivity. The multimedia
community nowadays has increasing interest in GIS applications, e.g., Flickr and Panoramio allow
users to upload images and videos with attached geo-locations of the cameras. Furthermore,
embedded sensors (e.g., GPS and compass units) have been cost-efficiently deployed on mobile
devices. Consequently, some very useful meta-data of videos, especially geographical properties of
video scenes, can easily be collected during video capturing. This association of video scenes and
their geographic meta-data has led to interesting research topics in the multimedia community. For
instance, the recorded sensor meta-data can be utilized to aid in modeling, indexing and searching
of geo-tagged videos at the high semantic level preferred by humans.
Current video search techniques that annotate videos based on the visual content are struggling
to achieve satisfactory results in online user-generated videos (UGVs), particularly in accuracy and
scalability. In this work, we utilized the camera's Field-of-View (FOV for short) to convert the
visual content into a series of spatial objects. Compared to visual content, the meta-data occupies
much less space, which makes searching among large scale of videos practical. To improve the
performance of our approach, a meta-data indexing algorithm is presented. Experimental results
show that the proposed methodology outperforms well in both effectiveness and efficiency.
4th International Conference on Computer, Mechatronics, Control and Electronic Engineering (ICCMCEE 2015)
© 2015. The authors - Published by Atlantis Press
875
资源评论
