Event Discovery in Multimedia Reconnaissance Data Using SpatioTemporal Clustering - PowerPoint PPT Presentation

Loading...

PPT – Event Discovery in Multimedia Reconnaissance Data Using SpatioTemporal Clustering PowerPoint presentation | free to download - id: 162d1c-Yzc5Z



Loading


The Adobe Flash plugin is needed to view this content

Get the plugin now

View by Category
About This Presentation
Title:

Event Discovery in Multimedia Reconnaissance Data Using SpatioTemporal Clustering

Description:

Bo Gong, Utz Westermann, Srikanth Agaram, ... During a reconnaissance mission, lots of data are collected in multimedia text, ... Mission event: Humvee ride ... – PowerPoint PPT presentation

Number of Views:38
Avg rating:3.0/5.0
Slides: 25
Provided by: ics5
Learn more at: http://www.ics.uci.edu
Category:

less

Write a Comment
User Comments (0)
Transcript and Presenter's Notes

Title: Event Discovery in Multimedia Reconnaissance Data Using SpatioTemporal Clustering


1
  • Event Discovery in Multimedia Reconnaissance Data
    Using Spatio-Temporal Clustering

Bo Gong, Utz Westermann, Srikanth Agaram, Ramesh
Jain Department of Computer Science University
of California, Irvine Irvine, CA 92697
2
Outline
  • Introduction motivation and task definition
  • Related work
  • Event discovery using spatio-temporal clustering
  • Event discovery from basic events
  • Spatio-temporal clustering
  • Experimental methodology
  • Dataset creation
  • Evaluation measures
  • Experimental result
  • Conclusion and future work

3
Motivation
  • DARPA EC-ASSIST Project
  • IBM
  • MIT
  • GATECH
  • AWARE TECH
  • UCI
  • During a reconnaissance mission, lots of data are
    collected in multimedia text, image, video,
    audio, sensor data.
  • Soldiers need to generate reports after a
    reconnaissance mission.
  • Event-based information organization can tell
    what happened during a reconnaissance mission.

4
Task Definition
  • Discover events that happened during a
    reconnaissance mission in multimedia data.

image
event
5
Challenges
  • Data are collected in an uncontrolled
    environment.
  • Quality of captured media may not be good.
  • The types of events to be detected are unknown in
    advance.
  • Events are to be detected in multimedia.

6
Outline
  • Introduction motivation and task definition
  • Related work
  • Event discovery using spatio-temporal clustering
  • Event discovery from basic events
  • Spatio-temporal clustering
  • Experimental methodology
  • Dataset creation
  • Evaluation measures
  • Experimental result
  • Conclusion and future work

7
Event Detection in Text (1)
  • Topic detection task in Topic Detection and
    Tracking (TDT) project.
  • The goal is to group together stories that
    discuss the same event.
  • Event-based topics are detected, e.g., Asia
    tsunami in 2004.

8
Event Detection in Text (2)
  • Event extraction task in Automatic Content
    Extraction (ACE) project.
  • The goal is to detect certain specified types of
    events that are mentioned in the source language
    data and recognize the selected information about
    these events and merge them into a unified
    representation for each detected event.
  • Pre-defined events are detected, e.g., Be-Born
    and Die are the event types.

9
Event Detection in Video
  • The events to be detected are predefined by
    applications.
  • Objects state change
  • A goal in a soccer game
  • Unusual events which rarely occur and are
    unexpected

10
Event Detection in Image
  • The goal is to organize images based on events.
  • Content-based or time-based.

11
Disadvantages of Existing Event Detection Methods
  • Most are content-based, which rely on media
    quality.
  • Utilize time information alone.
  • Many are developed for certain types of events.

12
Outline
  • Introduction motivation and task definition
  • Related work
  • Event discovery using spatio-temporal clustering
  • Event discovery from basic events
  • Spatio-temporal clustering
  • Experimental methodology
  • Dataset creation
  • Evaluation measures
  • Experimental result
  • Conclusion and future work

13
Mission Event and Basic Event
  • Basic Event a basic event is a media creation
    event or a detected event.
  • Mission Event A mission event is a higher-level
    incident occurring during a reconnaissance
    mission and is composed of semantically related
    basic events.

14
Detect Mission Events from Basic Events
  • The basic events have spatio-temporal
    concentration around mission events.
  • Discover mission events by clustering basic
    events.

15
Our Approach
  • Cluster basic events using spatio-temporal
    clustering techniques.
  • Time and location are independent of media.
  • Time and location are comparably reliable.

16
Spatio-Temporal Clustering (1)
  • Event similarity
  • where is the time difference
    between basic events
  • is the spatial distance between
    basic events
  • and are temporal and spatial
    thresholds.

17
Spatio-Temporal Clustering (2)
  • 1) Sort basic events in chronological order
    according to their start times.
  • 2) Form a cluster with the basic event which
    occurred the earliest.
  • 3) For each coming basic event i, compute its
    similarity with basic event i-1. If the
    similarity is 1, assign basic event i to the last
    cluster. Otherwise, form a new cluster with basic
    event i.
  • 4) Repeat step 3 until all basic events are
    finished.

18
Outline
  • Introduction motivation and task definition
  • Related work
  • Event discovery using spatio-temporal clustering
  • Event discovery from basic events
  • Spatio-temporal clustering
  • Experimental methodology
  • Dataset creation
  • Evaluation measures
  • Experimental result
  • Conclusion and future work

19
Dataset Creation
  • Obtained from a real DARPA evaluation with
    soldiers in a training area.
  • Two datasets captured by two soldiers.
  • Dataset 1 643 basic events, 7 mission events
  • Dataset 2 469 basic events, 6 mission events

20
Evaluation Measures
  • Evaluation metrics

21
Outline
  • Introduction motivation and task definition
  • Related work
  • Event discovery using spatio-temporal clustering
  • Event discovery from basic events
  • Spatio-temporal clustering
  • Experimental methodology
  • Dataset creation
  • Evaluation measures
  • Experimental result
  • Conclusion and future work

22
Experimental Results
23
Outline
  • Introduction motivation and task definition
  • Related work
  • Event discovery using spatio-temporal clustering
  • Event discovery from basic events
  • Spatio-temporal clustering
  • Experimental methodology
  • Dataset creation
  • Evaluation measures
  • Experimental result
  • Conclusion and future work

24
Conclusion and Future work
  • Mission events can be detected fairly well by
    spatio-temporal clustering on basic events.
  • The algorithm is far from the optimal. Better
    algorithms are needed.
  • When multiple soldiers data are combined.
  • When real-time detection is needed.
About PowerShow.com