多特征级联筛选的高鲁棒车牌检测

Robust license plate detection using multi-features cascading filtering

现有的车牌检测算法在车牌较模糊时往往难于取得很好的检测效果。针对监控图像的特点,首先提取清晰和模糊车牌所共有的归一化梯度特征,进行初步车牌检测;然后结合车牌区域的颜色直方图特征,进行级联筛选、去除非车牌样本,得到一种高鲁棒的车牌检测方法。基于真实监控图像的实验结果表明,此方法具有较高的稳定性和鲁棒性,尤其对模糊车牌具有明显优于已有方法的召回率。

All of the existing license plate detection algorithms can not get good detection results when the license plate is blurred. Considering the features of surveillance images, normalized gradient feature of plates in different clarity is extracted to do prelimina- ry detection, combined with the color histogram to make up a cascade construct in order to do further filtering and delete the sam- ples that are not plates, and get a robust license plate detection method. The simulating results with real surveillance images prove that the method has high stability and robustness, while it has a significantly better recall rate on blurred license plate than existing methods.