Enhanced Long-Range Computer Vision and Targeting

UAS are poised to play an increasingly prominent role in ISR and targeting applications. The Phase I effort made significant gains in demonstration of long-range computer vision technology; however, challenges remain. ATR technologies are generally limited to labeling target objects in images, one image at a time. Further, algorithm performance can vary significantly over a limited operating envelope characterized by a variety of factors including pose, sensor parameters, and target background. In this Phase II effort, we will address these deficiencies with respect to ISR and targeting applications. We will research, develop, and evaluate methods to reliably estimate target state using 2D image measurements; perform analysis to characterize algorithm performance sensitivities with respect to measurable and controllable factors; explore the use of synthetically rendered imagery to compensate for poorly represented view characteristics in available datasets; and develop approaches to exploit the resulting insights and/or performance models. To this end, we will explore platform behaviors, sensor configurations, and context-dependent algorithm parameters. The developed methods will represent "algorithm-aware" technology as a new concept in application of long-range computer vision. We will integrate these technologies into a developmental prototype system, and then conduct in-flight testing with increasing complexity to demonstrate capability.