NUS - VASuTool
VAST 2008 Challenge
Mini Challenge 4: Evacuation Traces
Authors and Affiliations:
Chidansh Bhatt, National
University of Singapore, chidansh@comp.nus.edu.sg
[PRIMARY contact]
Dhaval Patel, National University of Singapore, dhaval@comp.nus.edu.sg
[PRIMARY contact]
Mohan Kankanhalli, National University of Singapore, mohan@comp.nus.edu.sg
Wynne Hsu, National University of Singapore, whsu@comp.nus.edu.sg
Lee Mong Li, National University of Singapore, leeml@comp.nus.edu.sg
Student team:
YES
PhD. Supervisors : [Mohan Kankanhalli, Wynne Hsu, Lee Mong Li]
Tool(s):
We designed MATLAB based procedure for visualizing spatio-temporal trajectory for suspicious detections. We have provided required matlab files to reproduce the result. We are in process to build a more sophisticated automated suspicious detection tool with using provided matlab files. Readme file for this tool is available here. For novice user, please follow these steps as guideline to work with these methods.
Two Page Summary: YES
We have provided the detailed description. Actually we consider the competition as a part of our research idea. Our long term goal is to design an automated suspicious detection tool. Provided file explains the detail we have manually followed to obtain the results. We are implementing techniques discussed in paper for automated detection task. File is available here.
ANSWERS:
Traces-1 Where was the device set off?
We found a region near by which a device was set off. The cell is [65,31] near by which the device was set off. We encourage you to view this image.
Traces-2 Identify
potential suspects and/or witnesses to the event.
Potential
suspects and/or witnesses are people who were near the area just prior to
the explosion and exhibit suspicious behavior.
List of RFID tag numbers:
Suspicious:-
21 Ramon Katalanow [ Most suspicious : Reason: Visit the blast area and safely reach the exit area ]
29 Maxwell Lopez [ Less Suspicious : Reason: Speed was very high in the normal state, why?]
44 Karissa Graham [ Less Suspicious : Reason: Speed was very high in the normal state, why?]
Witnesses :-
56 Cleveland Jimenez [ We believe he is responsible employee so, he visit the affected area even after the blast]
28 Cecil Dennison [In evacuation state, this guy traverse near by region, so might be he know something]
1 Lindsey Bowles [In evacuation state, this guy traverse near by region, so might be he know something]
Traces-3 Identify any suspects and/or witnesses who managed to escape the building.
List of RFID tag numbers :
Suspicious:-
Witnesses:-
1 Lindsey Bowles [In evacuation state, this guy meet suspicious person in exit area]
Traces-4 Identify any casualties.
The people, who have not able to reach the safe region are victims.
Traces-5 Describe the evacuation
Office evacuation is the mass movement of persons from a dangerous place due to an explosive event. Ideally, an explosive event leads to drastic change in person’s movement. In other word, person’s movement was normal before explosive event taken place. After explosive events, every person starts running toward the exit (i.e., safe) area. Clearly, office evacuation scenario has following two states: (1) Normal State - a state before an explosive event and (2) Evacuation State - a state after an explosive event. For given dataset, it seems that most of people are not moving in normal state. But, after an explosive event, they run toward the exit area.
Plotting spatio-temporal trajectories in two dimensional space does not provide enough information during visualization. Hence, raw spatio-temporal data are transformed into suitable schematic which helps to address following issues:
1. Help to identify the approximate time during which an explosive event taken place (Answer of When?)
2. Able to differentiate trajectories which traverses the same location but at different time, i.e., fast trajectory and slow trajectory.
We defined two statistical measure named window based average speed and window based average traversed region to achieve these objectives and has following model to obtain the results.
Our proposed solution is simple and also easy to implement. We realized that time (i.e., temporal part) is equally important to the location (i.e., spatio part) in spatio-temporal trajectories specially for suspicious detection in office evacuation scenario. These is the reason why existing trajectory clustering based anomaly detection are not useful to discover suspicious trajectories. For detailed understanding of our approach we encourage you to read this report. Slide for this report is also available here.