A fuzzy logic automatic target detection system for LADAR range images

Abstract

This paper describes a f u m logic system that performs end-to-end Automatic Target Detection f o r LADAR imagery. Images are preprocessed with nonlinear f u q filters to remove noise. Several target detectors employing fuuy and robust statistical methods are used to create pixel-based target confidence values that are fused with the Choquet f u u y integral. Thresholds are determined from ROC curve analysis on a training set and then the detectors are run on a test set. Features are extracted from detected target windows and final target detection is peqormed by a neural network. Results on the training and test set are analyzed.

A Fuzzy Rule-Based Approach to Scene Description Involving Spatial Relationships

“…Basic vehicle detection algorithms are presented in [35]. Since SAM sites contain specific vehicle types, other recognition algorithms are necessary.…”

“…15 shows the testing on the SAM site LADAR image. The automatic target detection/recognition modules [35,36] worked very well here for segmenting and labeling the vehicles. The results from both the training image and the testing image describe each scene correctly.…”

“…Since the missile launchers display a rather unique shape, MSNNs are used in the SAM site scene analysis experiment to detect and identify missile launchers. If an object is not identified as a missile launcher, then features described in [35] are passed to a standard multilayer perceptron to make a decision on "center vehicle." The automatic detection and labeling results of Figs.…”

A new approach to target recognition for LADAR data

“…Other authors have developed ATR modules that are based on single frame images from the broadjump range channel data [6]- [8]. Hocaoglu et al [6] describe several nonlinear fuzzy filter designs used for detection.…”

“…Frigui et al [7] developed a third detection filter that is a robust variation of a constant false alarm rate (CFAR) design. Keller et al [8] use the detector suite developed in [3] and [4] as the first stage in a two stage detector. Input range images are preprocessed with an ordered weighted averaging (OWA, [9]) operator that replaces anomalous pixels and smoothes the remaining range values.…”