Multiple Sensor Fusion for Detecting Targets in FLIR (Forward-Looking Infrared) and Range Images.
Abstract
Automatic detection of tactical targets in corresponding sets of non-pixel registered forward-looking infrared (FLIR) sensor images and range sensor images was studied. A processing architecture was developed to address the problems associated with processing non-pixel registered imagery. The architecture used specialized sensor-dependent processing to segment the images, measure features, and analyze the single sensor feature data. The multiple sensor process of geometric registration, multiple sensor feature measurement, and multiple sensor target detection were then applied. Sensor-dependent segmentation processes passed a large fraction of the targets present in the imagery, along with a larger number of regions which did not correspond to any target. FLIR images were segmented based on pixel brightness. A new range image segmentation algorithm was developed which exploited the small-scale planarity of tactical vehicles. The post-segmentation target detection problem was that of partitioning segmented targets from segmented non-target regions. Feature information was processed to accomplish this task. The Bayesian minimum error criterion was adopted as the decision rule. When performance was optimized for all cases, the multiple sensor approaches were found to provide improved performance in all comparative performance measures. Theses. (jhd)
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 1989
- Accession Number
- ADA207577
Entities
People
- Michael C. Roggemann
Organizations
- Air Force Institute of Technology