Challenges with Per-Frame Metadata With Focus on Scalability
Video metadata opens up a lot of possibilities both for users and for content managers. Recently tools for automatically creating metadata via speech-totext,face recognition and object tracking among other techniques have made metadata even more relevant. Metadata for object tracking also creates th...
| Main Author: | |
|---|---|
| Format: | Others |
| Language: | English |
| Published: |
Umeå universitet, Institutionen för datavetenskap
2016
|
| Online Access: | http://urn.kb.se/resolve?urn=urn:nbn:se:umu:diva-128413 |
| Summary: | Video metadata opens up a lot of possibilities both for users and for content managers. Recently tools for automatically creating metadata via speech-totext,face recognition and object tracking among other techniques have made metadata even more relevant. Metadata for object tracking also creates the problem of needing to be bound to individual metadata frames. This thesis tries to find what the challenges are with per-frame metadata and how it could be stored in a way that scales vertically. A pre-study was made which tried to discover possible ways to deal with it and find pros and cons with each alternative. PostgreSQL (a relational dbms) was deemed the best alternative and the performance of it was tested by running a series of queries with different population levels. The results were that the search times seemed to be log(n) which means that it scales well. The relational database proved to work well in the other aspects as well. |
|---|