Developing Effective Strategies and Performance Metrics for Automatic Target Recognition

Abstract : University of South Alabama segment: In this report, we presented four multiple target tracking algorithms and two data/decision fusion algorithm for efficient target tracking in FLIR imagery. The performance of these algorithms has been evaluated using two approaches - evaluation based on the input scene data complexity, and evaluation based on the correlation output produced by each algorithm. Finally, we investigated target detection in the initial frame of a sequence using two techniques assuming no target information is known a priori. (Details included in the report). University of Memphis segment: We primarily focus on the performance measure characterization for both the dataset and our developed algorithms. We developed a composite metric table with different performance measures that demonstrates the capability of our two specific techniques, such as intensity and correlation algorithms, for detection and tracking. We also developed additional metric such as signal-to-noise ratio and classification for the entire dataset into low, medium and high categories. (Details included in the report). Wright State University segment: In this report, we explored the search engine design which allows for easy plug in of multiple search methods. Therefore, scenes can be evaluated based upon the performance of different matching algorithms. The key idea of this search method is to take advantage of the "divide and concur" concept. Instead of searching for a pattern in a large image, a smart approach is taken to divide the image space into overlapping pattern of sub-images. Search is then based on upon best match with sub- image. (Details included in the report).