红外前视对一类特殊建筑目标识别技术研究

Automatic Infrared Condensing Tower Target Recognition using Gradient Vector Features

针对电厂冷凝塔一类具有建筑规范的特殊建筑物,尝试在缺少其它基础地理数据保障的条件下,通过对冷凝塔外形的参数建模生成不同视点的参考图。在此基础上,针对红外/可见光多模图像匹配的特点,对自动目标捕获方案中模板匹配的相似性测度计算方法进行了研究。提出了基于梯度矢量相关系数的计算方法,将梯度信息作为红外/可见光多模图像的共性特征,匹配方案中不仅利用了图像点的梯度值,还充分考虑了该点的梯度方向信息。大量试验结果表明,该方法在识别性能上优于传统的灰度或轮廓匹配算法,匹配性能有较大的提高。

Condensing towers are such kind of special buildings whose shape can be well described using standard parameter equation according to corresponding architectural specifications.In order to perform automatic condensing tower target recognition,automatic target acquirement method based on template matching is proposed,that is to match real-time infrared images with visible satellite images.The template used here is generated only from condensing tower＇s parameter description.A gradient vector correlation method is also proposed to perform similarity measurement between infrared and visible images,which played an important role in automatic infrared targets acquirement.The proposed method takes gradient images of the multi-sensor images as common features.To avoid losing direction information,the image gradient field is considered as a two-dimensional vector matrix,and the gradient image＇s value and direction information are both used to computer correlation coefficient.Experimental results show that the proposed method achieves good performances for both day and night infrared images.