Paper Summary

1
Paper Summary 2By Jeff Ploetner

• Overview of Sensor Fusion Research at RDECOM
  NVESD
• Recent Results on Vehicle Detection Using
• Multiple Sensor Nodes
• Proceedings of the Sixth International Conference
  of Information Fusion, 2003

Philip Perconti James Hilger Science Technology
Division RDECOWCERDEC NVESD Fort Belvoir, VA,
U.S.A.
Murray Loew Department of Electrical and
Computer Engineering The George Washington
University Washington, DC, U.S.A.
2
Sensor Fusion Research

• Developed a flexible test bed for data collection
  and algorithm research
• Consists of multiple nodes and a base station
• Researching the best ways of fusing data
• No results given yet, just ideas and plans

3
Test Setup

• 3 nodes
• Sensors used
• Cameras (2 IR, 1 Visual)
• Microphone Array

4
Example Operation

• Use microphone array to detect direction of
  approach
• MUSIC algorithm determines angle of arrival
• Tell camera to track and take snapshots of the
  vehicle
• Extract image featuresfor further processing

5
Data Fusion

• Data can be processed locally or remotely
• Locally cuts down bandwidth
• Remotely can take other outputs into account
• Combine all data for better detection and fewer
  false alarms
• Cameras, Microphone, Weather, History
• Ground truth training data with GPS, manual
  labeling, vehicle drive-bys
• Looking for
• Car, Pickup Truck, SUV, Light Truck, Human
• Position, Direction, Velocity

6
Sample Outputs

• Data fused at base station can output more
  accurate measurements
• Multiple views can give better results

7
Example PerformanceEvaluation Metrics

• ROC Curves
• Receiver Operating Characteristics
• Tradeoff between detection and false alarms
• Confusion Matrix
• What gets confusedwith what

8
Chicago Pairing Surveillance Cameras with Gunshot
Recognition Systems

• The microphone system along with the camera can
  detect the sound of gunshots within a two-block
  radius, pinpoint where the shot was fired, turn a
  camera towards the shooter, and place a 911 call
  to the city's Office of Emergency Management and
  Communications.

9
ECE 285 Project

• Similar sensor fusion/detection/triggering
• Sensing options
• Video
• Standard as well as wireless
• Crossbow Motes
• Quad Passive IR Sensors
• Dual axis magnetometer
• Seismic Accelerometers
• Microphone arrays