Canny与改进Legendre矩结合的无人机边缘检测

UAV Edge Detection Using Canny and Improved Legendre Moments

摘要: 针对无人机机载平台计算资源有限，传统边缘检测算法存在精度不足、速度慢、边缘易断裂的问题，提出一种融合Canny算法与Legendre矩的亚像素边缘检测方法。该方法先通过Canny算法完成像素级边缘粗定位，再利用Legendre矩对边缘点进行亚像素精修，通过预生成并复用Legendre基函数模板减少冗余计算、降低复杂度。实验结果表明，该算法可有效提取无人机航拍图像完整边缘，定位精度达亚像素级，处理速度较Canny-Zernike算法提升约65%，计算量较传统Legendre方法降低50%~60%。该算法兼具高精度与轻量化、低耗时的特性可为无人机目标识别提供技术支持。

Abstract: To address the limitations of constrained computing resources on UAV airborne platforms and the inherent shortcomings of traditional edge detection algorithms—including insufficient localization accuracy, low processing efficiency, and frequent edge discontinuities—a subpixel edge detection method integrating the Canny algorithm and Legendre moments is proposed. Specifically, it first employs the Canny algorithm to perform pixel-level coarse edge localization, then applies Legendre moments to refine the detected edge points to subpixel precision. By pre-generating and reusing Legendre basis function templates, redundant computations are eliminated and the overall algorithmic complexity is significantly reduced. Extensive experimental results demonstrate that the proposed algorithm can effectively extract complete and continuous edges from UAV aerial images and achieves subpixel-level localization accuracy. Its processing speed is approximately 65% higher than that of the Canny-Zernike algorithm, and its computational load is reduced by 50%~60% compared with the conventional Legendre moment method. With the advantages of high accuracy, lightweight implementation, and low latency, the proposed algorithm can provide effective technical support for UAV-based visual target recognition.