Using Lifelines For SpatioTemporal Summaries

1
Using Lifelines For SpatioTemporal Summaries

• Anthony Stefanidis, Panos Partsinevelos, Peggy
  Agouris
• Dept. of Spatial Information Engineering
• National Center for Geographic Information
  Analysis
• University of Maine

2
Our Digital Government Project

• Title Knowledge Management Over Time Varying
  Geospatial Datasets
• Academic Partners
• University of Maine (lead)
• Penn State University
• University of California - Riverside
• Duration Aug. 2000 - July 2003

3
Agency/Industry Partners

• Direct
• National Imagery Mapping Agency (NIMA)
• National Agricultural Statistics Service (NASS)
• BAE Systems
• Indirect
• Open GIS Consortium (OGC)
• US Army Topographic Engineering Center (TEC)
• Federal Geographic Data Committee (FGDC)

4
Project Objective

• The development of novel meta-information
  structures conveying the content of time varying
  geospatial dataset collections, in order to
  improve access to and enable the analysis of such
  information

5
Paper Objective

• The automatic generation of video summaries.
• Providing support for
• spatiotemporal data reduction through the
  identification of important instances/frames
• spatiotemporal generalization using variable
  resolution as dictated by information content
• spatiotemporal behavioral analysis

6

• Lifelines S-T trajectories
• Need Generalization of S-T trajectories through
• selection of representative frames

7

• Object Trajectories
• as Paths in Multidimensional S-T Spaces

8
SOM Extraction of a Curvilinear Medial Axis
9
SOM Extraction (cont.)
Convergence
Ordering achieved
10
Input Imagery
Noisy MLC Results
HYDICE Test Area 1
11
SOM In SpatioTemporal Data

• S-T Nodes
• correspond to video frames (or datasets)
• of nodes generalization degree
• captures spatial and temporal behavior

12
S-T vs. S data

• S-T trajectories differ fundamentally from roads
  in that they are not as smooth or regular
• We need a better generalization approach to
  minimize information loss and redundancy

13
Standard SOM Generalization

• Problems
• Makes use of a preselected of nodes
• Handles poorly geometrically complex areas

14
A New Hybrid SOM

• Analysis of geometric complexity of generalized
  trajectory
• Introduction of mini-SOM solutions in complex
  data pockets
• Removal of redundant nodes
• Zooming-in in complex instances and zooming-out
  otherwise

15
Single object
Capture spatial and temporal variations turns,
ac/deceleration
16
Outline
t(2) t(1)
tracking
S-T trajectories
Area of interest
Proximity
Topology
Video dataset
SOM
Patches of moving object
Averaging, New Video
Node registration
Rough generalization
t
Thinning
t
x
y
Geometric-SOM generalization
S-T Summary
17
Geometry-SOM Technique
Thinning
Densification
18
Geometry-SOM
19
Generalization Detail
20
Accuracy Measures

• SOM vs. Hybrid

             
RMS(SOM) 89.3 RMS(Hybrid) 8.9
21
S-T SOM Nodes

• Instances where the object
• Accelerated/decelerated
• Changed orientation

22
Additional S-T Nodes

• Analyzing other spatial properties of
• single objects or relations among
• multiple objects
• Topology
• Proximity
• User-defined

23
Additional Nodes
Proximity
Reasoning
Area of Interest
24
Proximity Experiments
25
Data Handling

s (or t)
O1 O2 O3 O4
26
Lifeline Resolution
Attributes
Scale
Navigate through hierarchical structure in
information content and volume
27
Summary Generation

• Integrating S-T nodes
• SOM
• proximity
• reasoning
• Registration and grouping
• as average (titj)/2
• as video segment from t(I) to t(j)

28
Supported Queries
How similar are the S-T trajectories of two
objects?
Total Attribute Similarity (Partial Attribute
Similarity Indices)
3-D similarity
29
Find all objects for which a specific attribute
varied in a specific way over a certain extent.
Query increment
O1
O2
Object1 53 Object2 75
30
Scale Independent Comparisons

• Differential registration
• Lifeline stretching to reveal hidden relations.

31
Event Similarity Matching

• Crime rate
• Pollution

Urban Growth
Relations
32
To Be Continued
For more information http//www.spatial.maine.
edu/peggy/dgi.html or tony_at_spatial.maine.edu