【摘要】 提出一种基于显著性特征的交通信号灯检测方法,并通过SVM分类器进行识别。首先,对低分辨率图像生成颜色、亮度和边缘特征图并融合成显著图。其次,获取交通信号灯候选区域,利用交通信号灯的几何特征和颜色特征过滤噪点。然后,提取交通信号灯目标,得到只包含单个箭头灯、圆灯和数字灯的区域。最后,提取检测区域的HOG特征,通过SVM分类器进行识别。实验结果表明,该算法检测率和识别率均超过97%。更多还原

【Abstract】 An approach for detecting traffic lights is proposed based on salience feature,and recognized by SVM. Firstly,col-or,luminance and gabor edge feature map are generated in low resolution resolution of image,and merged into saliency map. Sec-ondly,traffic lights candidate region is obtained,and noise is filtered according to geometric and color features of traffic lights.Thirdly,the objects of traffic lights are obtained,the areas which include only one arrow,circle and digit lights are gotten. Lastly,HOG feature and recognizing are extracted by SVM. Experiment results indicate the detection rate and recognition rate of the pro-posed method are over 97%.更多还原