Summary of talk

• Context: DERA Event Stream Analysis Project
• Problem: Indexing large data sets
• Solution: Translate simulator logs to VRML2.0

Context: Event Stream Analysis

• Concept: event driven systems generate massive amounts of data continuously
• At present we throw most of this data away
• ESA concept: "mine" this asset

Potential uses

• validation and verification
• post exercise review
• training and assessment
• anomaly detection
• system modeling

ESA project threads

• Visual programming and visualisation languages/tool kits (PUMBER)
• Machine learning/prediction techniques
• Visualisation techniques
• Indexing techniques
• Case Studies
• Exemplars

Plumber

The Indexing problem

• Simulators generate large data sets: how can the analyst identify interesting sub-sequences for detailed analysis?

• Three approaches:
• Human analyst searches through large printout of raw data
• Syntax directed pattern search
• Make a movie from the data?
• But there are lots of ways of making movies from data?

Approaches to making movies from a dataset

• ARTIFICIAL MAPPING: map the data onto some N-dimensional space, and then provide tools for 'flying through' projections of this space
• not obvious which mappings to choose
• needs to be explained to customer

• NATURAL MAPPING: use the semantics of the application

• e.g. for combat simulators, every event has a position in the usual 3D space + time dimensions
• but not all datasets have such easy mappings

Do we need new technology?

• Most combat simulators already have replay/cue and review facilities
• However, these are used only rarely by analysts:
• can't easily be accessed on the analyst's laptop
• expensive resource
• limited functionality
• We seek a 'cheap and cheerful' technology which will run on a laptop PC

• Can't afford to develop the rendering technology
• Can't afford to develop complex dataset->movie technology
  
  
• Goals:
• both translator and replay software should run on a laptop
• use standard components wherever possible
• should connect to other analysis tools (e.g. plumber, object databases, detailed terrain rendering tools, etc.)

 Progress to date

• 3Q96:
• requirement identified
• VRML2.0 standard targeted

• 4Q96:
• translation techniques for simulator dataset to VRML2
• first runs with SGI Cosmo player on Indy

• 1Q97:
• fast forward / replay pause mechanisms

Future work

• Move to PC technology as soon as a stable PC VRML browser appears
• Connect to:
• object databases
• terrain databases
• plumber and hence other tools

• Work needed on
• HCI aspects, using real analysts input

 References

• Visualisation Techniques for Event Stream Analysis, Sue Haines, Richard Kennaway, Ronan Sleep, Proc. EuroGraphics9UK97.

• Making Movies from Simulator Logs using VRML2.0, Sue Haines, Richard Kennaway, Ronan Sleep, UEA report April 1997.

• Demo video in production?