Subject: RESCUE MEETING
Date: 01/23/04
Loc: URP
Author: AM
Discussion of imminent visit by sponsors. Outline of story and demos to be built in preparation.
SM: *Site visit March 19. “Pre” site visit.
*By March 1: A set of demos.
Darleen Fisher, PM, RI.
Bhavani – her boss.
9am-5pm all day meetings.
*Assign handful of students to meet with them.
Review proposal:
UCSD is ahead of us. (schedule-wise)
Need to match their story.
Primary goal: Research infrastructure grant (equipment)
[White board]
[Inner circle: people with 100 handhelds.
Outer circle: Videocams paired with network equipment.
Rescue office: Service, store data.
Sync with UCSD campus.
Mobile cyber-shuttle.]
Hi end vs low end cameras – pan, zoom, etc.
Need to follow the script on this.
What will we demo.
CAMAS – now a few months old.
What’s doable in 1.5 months?
Ganz looked at our demo.
- Integrated with voice-activated system.
Navigate problem tree with voice commands.
NV: *Borrow voice stuff from Ganz.
SM: *Dmitri* talk to Ganz.
NV,SM: More UCSD-UCI coordination.
SM: *Use people 100% on Rescue. Dmitri, Amnon.
*Campus-level real data.
*Facilities (Greg Fountain), NACS (Dana Rude), Police, Parking
*AM* move this forward. Next week get people in same room, in Cal-IT.
*Within 1 week, get at least one data set.
Data source: bulding plans.
*Ram* talking to them.
Geo-register with PDAs. Phone & cam.
*Buy 1 or 2 PDAs.
NV: *Get cost.
CB: Speed of algorithms.
- Take hours, may not get to RT.
SM: Canned.
Before & after.
NV: Localize in space and time.
CB: *CB* be in on meeting with UCI people.
All/Chen: Blue sticks/emergency…
CB: Need to simulate emergency?
CERT = Community Emergency Response Team.
*Leverage them.
SM: ? has a team of CERT people.
Agenda for social science.
Richard Matthews – Group is ?
NV: Tie in Padhraic’s stuff.
Sridevi: Trajectory models.
SM: *Integrate into CAMAS story.
CB: Modify to do group size detection.
SM: Learning a model of occupancy.
CB: Condition on occupancy.
Tells start and end.
Piecewise Brownian motion…
Story more important than demo.
Sridevi: Prerecorded data.
SM: Trajectory analysis.
10 x 30 second segments of video.
NV: Tie into reliability model?
CB: To do anomaly detection.
SM: Some source > analysis > event extraction.
All: Demo first vs story first.
CB: Use augmented spaces.
Additional to physical space.
Off the shelf technology.
Easy to demo.
Other patterns, associations.
Eg, types of system failures.
SM: Improve the CAMAS demo.
Add real data set.
Make it student oriented?
Sororities, fraternities, clubs.
CB: UCSD did some stuff.
SM: * [CAMAS1 = DEMO #1]
*March 1 demo system.
Camas was mainly an analysis demo.
*TEAM:
Dmitri – leader
Amnon
Ram
Haimin
Jehan (1/2 ?)
SM: * [DEMO #2]
*[PRIVACY PRESERVING DATA COLLECTION]
(Video surveillance)
*Livewave.com
Video cam for bio, chem. Surveillance.
(Multi-source validation)
Canadian – monitor hockey games…
?: RF ID tag for tracking.
SM: Merge models.
Jehan: Equipment arrives next week.
*Need Cal IT to submit…
SM: *Kaushik worked on a 3 VC system.
Jehan: One open area, one room.
Iosif: *Find cam system.
SM: Hide identity, till someone breaks a rule.
Story of someone fired for too many coffee breaks.
GS: Nitesh…
SM: Cryptography.
GS: BYU, derivatives of Trust-Builder.
Go ask Magda?
SM: *TEAM:
Jehan – leader
Mahesh
Nitesh
ImageCat – vidproc experience?
SM: *[DEMO #3]
*[ALTERNATE 911 SERVICE]
High call volume.
Dynamic analysis.
Disambiguate calls.
Build larger model.
Same event or not.
Set of questions to ask.
Yiming – working filtering.
Would fly well with response community.
[Event E1 > S, T, P1, P2, P3. E2, E3 …]
Space, Time, Property1, Property2, etc.
Call > Compare to existing events > Extract.
If 95% accuracy, major value added, hot stuff.
NV: Classification problems.
Can do some of this a priori.
SM: Forget voice for now.
Extract from NLP?
CB: Examples of transcribed 911 calls.
Some available in news articles.
News: accident logs.
? websites.
SM: Claim is that a data-driven approach is better than the complex NLP done up to now.
YM: Event generation.
SM: Information extraction expertise – Amnon.
Alternate 911 is part of analysis and store [Rescue Office, in earlier diagram].
YM: [Argument for YM’s project, approach.]
Assume features are already extracted.
Automatically prioritize, rank calls.
Given type of emergency.
Relative priority of calls coming in.
SM: Triage to individual = OK.
Personalize.
Filter.
Why real time?
From speech?
Start with keyword extraction.
Space, time, type.
NV: Tie to Ganz notification work.
Collection, analysis, dissemination.
[Debate with SM]
SM: *TEAM:
Yiming
Qi
Kemal
Chen?
Amnon?
NV: 4th loop sensor data?
All: Discussion of similarity of CAMAS vs. Alternate 911.
NV: *[DEMO #4]
*[COLLECTION DEMO]
Sensor networks.
Live loop sensor data.
Show map, data coming in.
Display data.
Display collection.
Cost-effective real-time policies.
Xingbo: [Presentation: Berkeley Traffic Management Ctr Website]
http://www.pems.eecs.berkeley.edu
22,000 sensors deployed on all California highways.
TMC = Traffic Mgmt Ctr.
4 levels:
(4) TCM
(3) Loop detector
(2) Vehicle station
(1) Sensor level
+ Link communications
Vehicle speed:
(1) by multiple detectors
(2) by occupancy and flow.
Trafficdodger.com
SM: 2GB/16 hours.
All: Discussion of data demo, new angle on this traffic data.
SM: [Presentation: Earthquake before and after slides, etc]
Video could be used to do large-scale assessment in real time.
All: Note team assignments above, get demo work going.
Note starred items and others above.
AM: Set up meeting with UCI facilities, etc.
Dmitri: Talk to Ganz.
Ram?: Buy PDAs.