Hardware Software Co-Design for Real Time Detection of Small Target in IR Video

For Infrared Search and Track (IRST) systems used in military applications, presence of small target in a large background is quite common. To detect and track those objects in presence of background clutter and noise is very difficult especially when full field of view processing is employed. Sophisticated computer vision techniques perform well but with an expense of heavy computational cost. In this paper, we have tried to solve these problems in two ways. First, a hardware software co-design method employing FPGA is used to accelerate the search of small object using local contrast variation within a sliding window. Second, the location cue is taken from hardware board and a Region of Interest (RoI) based software processing is employed to segment and enhance the target and suppress the background clutter and noise. A fusion of edge detection (linear predictive filter based) and saliency map is employed to detect and enhance the target. We compared our results with baseline detectors (generally used for infrared (IR) target tracking) in terms of timing and detection accuracy. In both fronts, our method gives better performance than the conventional methods with relatively high performing methods like Improved Local Contrast Measure (ILCM), Markov Random Field (MRF) and Otsu Threshold with Connected Component Labeling (CCL).