FLIR Image Analysis with the Autoscreener Computer Simulation. Volume I.
HONEYWELL INC MINNEAPOLIS MINN SYSTEMS AND RESEARCH CENTER
Pagination or Media Count:
The final report presents the results of a recent investigation to assess the feasibility of automatic target cueing technology in detecting and recognizing tactical targets in FLIR imagery. The FLIR imagery data in digital form as well as in transparencies was furnished by the U.S. Army Night Vision Laboratory. The digital data was processed by Honeywells digital computer simulation of the Autoscreener hardware. The digital processing extracts the first level features edge and contrast information from which object intervals and candidate objects are extracted. Each candidate object is represented by a vector with eighteen components based on the object shape and its texture. The classifier design was investigated next. Based on statistical methods ten features were selected one set for large objects and another for small objects. The largesmall object separation was based on size. In addition a training and test set was selected. The classifier for each case was trained by generating six discriminant functions, the decision space and training confusion matrix. The classifier was then tested on the remaining set test set and the confusion matrix was generated and evaluated using ground truth. The overall system as well as the classifier detection, recognition and false alarm probabilities were estimated.
- Infrared Detection and Detectors