Drone detection in images is a critical task for security and surveillance applications. Traditional thresholding methods are commonly used for contrast-based object detection but often suffer from high false positives, low precision, or computational inefficiency. In this paper, we propose a novel Fourier-based adaptive thresholding algorithm that utilizes Fourier transform to approximate the image histogram, enabling robust identification of multi-modal background distributions for precise threshold selection. The method introduces several key innovations: 1) low-pass filtering of the histogram via FFT to suppress noise while preserving dominant intensity modes, 2) multi-peak detection for handling complex outdoor backgrounds with sky, clouds, and terrain, 3) explicit overexposure attenuation to mitigate bright sky effects, and 4) computational efficiency through histogram-based processing independent of image resolution. We evaluate the proposed method against 12 existing thresholding methods on 116 benchmark images, demonstrating superior performance with precision of 81.4% (11.66% higher than the nearest competitor), recall of 94.0% (11.37% improvement), and coverage of 61.4% (48.31% improvement). The method achieves processing times of 0.817 ms at 4K resolution ( 3840×2160 ), delivering 27.4× speedup over the nearest quality competitor while maintaining exceptional detection accuracy, making it ideal for real-time drone surveillance in resource-constrained embedded systems.
