一種用于復雜工業(yè)場景的相機標定方法

打開文本圖片集

中圖分類號：TP391.4 文獻標志碼：A 文章編號：2095-2945（2025）18-0071-04

Abstract：Circularpatternswithencoded informationare favoredincameracalibrationduetotheirhighaccuracyand stability.However，incomplexindustrialenvironments，calibrationboardsmayfaceisuessuchasrotationandoclusionand traditionalglobalthresholddetectionmethodsarepronetofailure，leadingtocalibrationfalures.Therefore，anewmethod combiningdeplearmingwithedgesubpixeldetectionisproposed.Thismethodfirstdetectstheminimumcircumscribedframeby rotatingtheboundingbox，thenusesaffnetransformationtoeliminateecentrcityerrors，andfinallyuseslocalthresholdsto detectedgesub-pixelstoachieveacuratecalibration.Experimentalresultsshowthattheproposedmethodismoreeficientand robust than traditional methods.

Keywords：camera calibration; coded target;rotated bounding box;deep learning;edgesubpixel detection

計算機視覺的三維測量因非接觸、高精度、快速響應和強適應性而廣受研究與應用[-]。（剩余4675字）