铁路场景下的红外与可见光图像自动配准算法

Automatic registration algorithm of infrared and visible images in railway scene

现有基于机器视觉的铁路异物侵限检测方法大都是利用单一可见光图像进行处理。针对其在夜间及恶劣天气条件下因图像视觉效果变差而导致异物准确报警率不高的问题,提出了铁路场景下的红外与可见光图像的自动配准算法。针对异源图像配准中误匹配率高的问题,提出了基于SURF特征点检测、提取和匹配的改进算法,先对初始同名点进行几何约束条件下的筛选,然后利用图像结构相似度理论进一步剔除外点,最后用RANSAC算法进行最终的精匹配。实验结果表明,该算法能实现图像的自动准确配准,处理后的图像全面体现了图像温度、色彩轮廓细节等信息,更利于后续的目标识别和异物准确报警。

In existence,most methods based on machine vision for detecting the foreign object intrusion are using a single visible source vision to process.Aiming at the problem that the image visual effect will be affected in the night and poor light condition,which leads to the low alarm accuracy of the foreign object,an automatic registration algorithm for infrared and visible images is proposed.Pointing at the issues of high false matching rate in the registration of multi-modal images,An improved algorithm for for feature points detection,extraction and matching based on SURF is proposed.Firstly,the initial homonymous points are filtered by the rule of geometric constraints,and then the image structure similarity theory is used to further eliminate the outer points.Finally,the RANSAC algorithm is used for the final accurate matching.The experimental results show that the algorithm can achieve accurate and automatic image registration and the processed image fully reflects the image information about color and temperature and outline details which is beneficial to the subsequent target recognition and accurate alarm of foreign objects.