Adaptive detection threshold selection for vision-based sense and avoid

Current state-of-the-art vision-based sense and avoid systems based on morphological and hidden Markov model filtering require the manual selection of static (time-invariant) detection thresholds. Manually selecting suitable static detection thresholds is challenging (and currently requires highly trained operators) because it involves balancing tradeoffs between detection and false alarm performance in different image sensing conditions. In this paper, we exploit recent work on the characterisation of vision-based aircraft detection problems in the sky-region to propose an adaptive threshold selection approach. Using data sets captured during flight experiments, we show that our proposed adaptive threshold approach can enable improved detection range performance compared to manually selected static thresholds.