Here is what we have done in this study:
1). Our previous results of spatio-temporal fusion for target classification have been further developed for target detection. (SPIE AeroSense, Vol. 4731, pp. 204-215, April, 2002)
2). Different temporal integration (fusion) strategies have been developed and compared, including pre-detection integration (such as additive, multiplicative, MAX, and MIN fusions), as well as the traditional post-detection integration (the persistency test).
3). In our 2nd study, The temporal correlation and non-stationary properties of sensor noise have been investigated using sequences of imagery collected by an IR (256x256) sensor looking at different scenes (trees, grass, roads, buildings, etc.).
4). The natural noise extracted from the IR sensor, as well as noise generated by a computer with Gaussian and Rayleigh distributions have been used to test and compare different temporal integration strategies.
Some preliminary results are summarized here:
1). Both the pre- and post-detection temporal integrations can considerably improve target detection by integrating only 3~5 time frames (tested by real sensor noise as well as computer generated noise).
2). The detection results can be further improved by combining both the pre- and post-detection temporal integrations.
3). The sensor noise at most (> 95%) of the sensor pixels are near stationary and un-correlated between pixels as well as (almost) un-correlated across time frames under a good weather condition.
4). The noise at a few pixels near some surface edges has shown non-stationary properties (with increasing or decreasing mean across time).
Concealed weapons pose a significant threat to military and civilian personnel protecting secured facilities and in low intensity conflicts. Passive millimeter wave and highly sensitive infrared sensors can detect these weapons. Parallel processors and a sensor fusion algorithm developed for engagement of military targets promise to solve this problem.
Alternative algorithms for detecting and classifying mines and minelike objects must be evaluated against common image sets to assess performance. The Khoros Cantata<SUP>TM environment provides a standard interface that is both powerful and user friendly. It provides the image algorithmist with an object oriented graphical programming interface (GPI</SUP>. A Khoros user can import 'toolboxes' of specialized image processing primitives for development of high order algorithms. When Khoros is coupled with a high speed single instruction multiple data (SIMD) algorithms. When Khoros is coupled with a high speed single instruction multiple (SIMD) processor, that operates as a co-processor to a Unix workstation, multiple algorithms and images can be rapidly analyzed at high speeds. The Khoros system and toolboxes with SIMD extensions permit rapid description of the algorithm and allow display and evaluation of the intermediate results. The SIMD toolbox extensions mirror the original serial processor's code results with a SIMD drop in replacement routine which is highly accelerated. This allows an algorithmist to develop identical programs/workspace which run on the host workstation without the use of SIMD coprocessor, but of course with a severe speed performance lost. Since a majority of mine detection componenets are extremely 'CPU intensive', it becomes impractical to process a large number of video frames without SIMD assistance. Development of additional SIMD primitives for customized user toolboxes has been greatly simplified in recent years with the advancement of higher order languages for SIMD processors (e.g.: C + +, Ada). The results is a tool that should greatly enhance the scientific productivity of the mine detection community.
A digital orthophoto can be obtained by rectifying a digitized perspective aerial photo. The orthorectification requires a terrain elevation model to remove the displacement caused by terrain relief. This paper describes a technique for partitioning a large photo into multiple subimages, mapping them onto the array processor for rectifications, and combining the results into an orthophoto. The combination of innovative algorithms and the massively parallel processor architecture leads to a significant improvement in throughput of digital orthophoto production.
Real-time image segmentation is a critical part of an automatic target recognizer (ATR). The segmentation of poorly resolved targets in low contrast thermal video images is a challenging task. Most edge-based segmenters are too susceptible to noise, contrast variation, and target boundary discontinuity. The problem is the lack of a fast (video rate) and robust method of grouping the relevant edge elements together while rejecting the irrelevant ones. We have overcome this problem by combining the processing power of a single instruction multiple data (SIMD) computer with a newly developed model-directed segmentation algorithm.
Although there are many edge detection operators that are used in Automatic Target Recognition (ATR) Systems we believe that the required performance can be accomplished using simple operators such as the Sobel operator with minor modification through pre and postprocessing. In this paper we describe two methods that enhance the Sobel performance. The first which can be considered pre processing increases the effective size of the operator to 5*5 The second which is postprocessing modifies the edge magnitude output based on the consistency of the local edge direction. Both methods can be easily implemented on SIMD machines and they are effective in deleting isolated edge points which usually are not part of interesting targets. We will describe the implementation of these techniques on a SIMD machine and study their effect on the performance of an ATR system. 1.