CAJ | 학술논문

变化光照条件下交通标志检测算法的准确率往往会显著降低。针对此问题，提出了1种新颖的概率图建立方法，并结合最大稳定极值区域特征进行交通标志检测。该方法包括3个处理步骤：①根据不同光照条件对真实场景交通标志样本图像进行明确分类以构建多类颜色直方图，将交通标志输入图像由原始色彩表达转变为概率图（直方图反投影）；②通过在概率图上进行 MSER特征提取，获取候选的交通标志区域；③根据候选区域的面积、宽高比等特征快速有效去除非交通标志区域。实验结果表明在弱光照和强光照条件下基于归一化RGB的交通标志检测算法检测准确率分别下降到84．4％和83．0％，基于红蓝图的交通标志检测算法检测准确率分别下降到87．4％和86．3％，提出的算法在变化光照条件下依然可以保持90％以上的检测准确率，对光照变化有较好的鲁棒性。
변화광조조건하교통표지검측산법적준학솔왕왕회현저강저。침대차문제，제출료1충신영적개솔도건립방법，병결합최대은정겁치구역특정진행교통표지검측。해방법포괄3개처리보취：①근거불동광조조건대진실장경교통표지양본도상진행명학분류이구건다류안색직방도，장교통표지수입도상유원시색채표체전변위개솔도（직방도반투영）；②통과재개솔도상진행 MSER특정제취，획취후선적교통표지구역；③근거후선구역적면적、관고비등특정쾌속유효거제비교통표지구역。실험결과표명재약광조화강광조조건하기우귀일화RGB적교통표지검측산법검측준학솔분별하강도84．4％화83．0％，기우홍람도적교통표지검측산법검측준학솔분별하강도87．4％화86．3％，제출적산법재변화광조조건하의연가이보지90％이상적검측준학솔，대광조변화유교호적로봉성。
Aiming at the problem that the accuracy rate of traffic sign detection will become significantly lower in protean illumination scenario ,a novel robust method of traffic sign detection is proposed based on the color probability map which is built from multiple color-histogram back-projection and the extraction of MSER (Maximally Stable Extremal Region) in color probability map .The algorithm consists of three steps :1) Sample images of traffic signs are classified into a series of different subsets with different illumination states for each color of interest (red ,blue or yellow ) and the color probability map is built from the multiple color-histogram built from each subset of sample images ;2) Candidate re-gions of traffic sign are found by using the extraction of MSER in color probability map ;3) Non-traffic-sign regions are e-liminated efficiently according to the features (region perimeter ,area ,etc .) of the detected MSER .Experimental results show that under the conditions of low light and strong light the accuracy rate of traffic sign detection algorithm based on normalized RGB drops to 84 .4% and 83 .0% respectively ,while the accuracy rate of traffic sign detection algorithm based on red/blue image drops to 87 .4% and 86 .3% respectively .The proposed method can still remain more than 90% of the detection accuracy in protean illumination scenario ,and is of higher robustness in protean illumination environment .