A Fast Working System for Tracking Multiple Objects in a Confined View Space
Stan Sexton, UNR Department of CS.
Dr. Jim Gattiker, Los Alamos National Laboratory
Dr. George Bebis, UNR Department of CS.
Dr. Dwight Egbert, UNR Department of CS.
Home
Abtract
Problem Statement
Results
Methodology
Conclusion & Future Work
Acknowledgements

Methodology


Main Steps

  1. Background segmentation
  2. Location of Objects
  3. Recognizing Object Events
Background segmentation

A system of equations must be set up for every pixel within the images. For each pixel we can compute the mean and standard deviation over time. The use of these two values will then tell which pixel values are significantly different than the background model. Thus we have a binary image containing white foreground pixels and a black background pixels.

Location of objects

A breadth first search is done on the pixels to classify them into blobs.

Recognizing Object Events

For this project I concentrated on the following object events. By no means is this a complete list of available object events and should not be taken as such. These are the events that I found to be most interesting for the current project.

  1. Constant momentum
  2. Stopping:
  3. Movement towards or away from the camera
  4. Arrivals
  5. Departures
  6. Occlusions

UNR Homepage | UNR Department of Computer Science | UNR-CVL | UNR CRCD | LANL Homepage | LANL NIS-7 | NSF

Valles Caldera National Preserve Valles Caldera National Preserve Valles Caldera National Preserve