基于梯度信息融合的海面红外目标实时检测

Real-time detection of marine infrared objects based on the fusion of gradient information

针对海面红外序列图像的特点,提出一种适用于海天背景和海岸背景的目标检测方法。该方法不需要进行特殊的预处理来滤除图像噪声,首先量化子图像的区域复杂度以及单元区域上下邻域的灰度差异,从而预测出海界限区域的位置;然后利用改进Canny算子提取轮廓边缘,突出海界线特征,降低其他纹理的干扰,再进行Hough变换,提取出海界线;最后提出泛化梯度概念,通过融合的梯度信息,联合海界线的位置,进行海面目标标记。实验表明,所提方法能准确地定位海界限区域以及水平或倾斜的海界线,并快速精确地提取红外海面目标。整个过程的平均花费时间约为60 ms,具备很好的鲁棒性和实时性。

A feasible method considering the character of marine infrared image was proposed to detect objects in the sequential images from surface vehicle, which was not only appropriate for sea-sky background but also for offshore background. There was no need to filter the noise. Firstly, the complexity of sub-images and the average gray difference of their up and down neighborhood were measured to predict the sea line region. Secondly, improved Canny edge detection was applied to extract the contour of the sea line region. It made the sea line obvious and meaningless edges disappear greatly. Thirdly, Hough transformation was used to pick the longest line as the sea line. Finally, a kind of general concept of gradient was put forward. The targets could be marked excellently under the fusion of gradient. The experiment results show that this method can locate the sea-line region and the sea line with any tilt reliably and obtain the information of objects effectively. The whole procedure costs about 60 ms and it is real-time and robust.