Upload
ellen-barton
View
224
Download
0
Embed Size (px)
Citation preview
Performance linked Workflow Composition for
Video Processing – An Ecological
Inspiration
Jessica Chen-BurgerJessica Chen-BurgerUniversity of EdinburghUniversity of Edinburgh
An Ecological Motivation An oil spill occurred at Lungkeng near
Ken-Ting ( 墾丁龍坑生態區 ) the head of the Environmental Protection
Administration (EPA), Lin Jun-yi vowed to restore it to its former condition within 2 months.
But it is unclear as how this may be achieved –
There was no prior survey on the area - there isn’t a basis for referring to Lungkeng's original ecosystem prior the oil spill.
Source: Taiwan News, http://www.etaiwannews.com/Viewpoint/2001/02/14/982136471.htm
In addition, if there was such research data into the area's ecology before the spill, one could have used it as a basis to seek insurance compensation !!
In Response In 1992, TERN (Taiwan long-term
Ecological Research) project, a join effort with US NSF long-term ecological research, were formed.
Sponsored by Taiwanese National Science Council (NSC).
Wireless Sensor Nets were constructed and managed by NCHC.
NCHC (National Center for High-performance Computing).
Sensor Grid in Taiwan
福山
關刀溪
鴛鴦湖
南仁山
塔塔加
Ken-Ting coral reef at Third Nuclear Power Station Adapted from Source: NCHC
墾丁 Ken-Ting National Park
Under-water surveillance
Objectives and Scopeof EcoGrid
• To develop a scalable observational environment that is capable to hierarchically connect local environmental observatories into a global one via grid and web-service technologies.
• To enable scientific and engineering applications in long term ecological Research (LTER) as well as environmental hazard mitigation.
• To provide an end-to-end process from automatic information collection to automated analysis and documentation.
• To provide a useful feedback loop for Ecologists. • Relevant Technology and solution:
• Self-aware and adaptive workflow composition and management.
Challenges The vast amount of data available to us is of
tremendous value. However, how to process them efficiently and
effectively is a big challenge: – One minute of video clip takes 1829 frames and
3.72 Mbytes;– That is 223.2 MB per minute, 5356.8 MB per day,
and– 1.86 Terabytes per year for one operational camera; – Currently there are 3 under-water operational
camera.
Human Efforts:– Assuming one minute’s clip will need one human expert
manual processing time of 15 minutes: – This means that for one camera and one year’s recording
will cost a human expert 15 years’ efforts just to do some basic annotation work;
– This is a hopeless situation and automation must be deployed in order to carry out these tasks efficiently and effectively.
In addition, relevant clips need to be related, organised, classified in a sensible structure, and so that additional properties may be further derived, however, this is again time consuming.
Challenges
Dynamic nature of collected video Target information is variable and un-
predictable Limited expertise Untrained Grid/workflow tool users
Challenges
Effective and efficient workflow automation
Data co-relation identification, management and retrieval
Presentation of information– Rendering of images– annotation – co-relation with other information/clips
Opportunities
Rich processing opportunity Long-term ecological documentary and
analysis Flexible practice that is incrementally
improved over time Semantic based annotation