Developing Effective Strategies and Performance Metrics for Automatic Target Recognition

The final report that summarizes the work performed at University of South Alabama (USA). Four different target tracking algorithms and two data fusion algorithms have been developed for single/multiple target detection and tracking purposes. Each tracking algorithm utilizes various properties of targets and image frames of a given sequence. The data fusion algorithms employ complementary features of two or more of the above mentioned algorithms. Thus, the data fusion technique has been found to yield the best performance. The performance of the above mentioned algorithms was evaluated using two approaches evaluation based on the input scene data complexity, and evaluation based on the correlation output produced by each algorithm. We developed two new performance metrics to evaluate the effect of input plane data complexity on the performance of the algorithms. To evaluate the output produced by each of the aforementioned algorithms, we used four performance metrics, such as peak-to-sidelobe ratio, peak-to-correlation energy, peak-to-clutter ratio, and Fisher ratio. Finally, we investigated the problem of target detection in the initial frame of a video sequence using two techniques, namely feature vectors and multilevel data fusion, assuming that no target information is known a priori. 1. Multiple Target Tracking Algorithms We developed four new multiple target tracking algorithms which are based on fringe-adjusted joint transform correlation (FJTC), intensity variation function (IVF) and template matching (TM) techniques. The performance of these algorithms were tested using 50 real life forward looking infrared (FLIR) image sequences supplied by the Army Missile Command (AMCOM). 1.1 FJTC, IVF and TM (FJTC-IVF-TM) Based Multiple Target Tracking In this section, we introduce FJTC-IVF-TM based multiple target tracking algorithm which includes frame preprocessing, motion estimation and target tracking. In the initial stage, we utilized the FJTC technique [1]. Using FJTC results, the current frame of the image sequence is recovered after global motion compensation. For the tracking algorithm, IVF and TM techniques have been used. The IVF based target tracking approach primarily utilizes target intensity information associated with the previous frame and the current frame. Initially, a target reference window and a subframe are segmented from consecutive frames. Then the IVF is generated by sliding target window inside the frame or the subframe, which yields maximum peak value for a match between the intensity variations of the known reference target and the unknown candidate target window in the frame or the subframe as shown in Fig.1.