基于显著性与稠密光流的红外船只烟幕检测方法研究

Ship smoke detection method based on saliency and dense optical flow

船只目标是海上的一种重要目标。红外成像系统由于其可白天、夜间同时工作的特点已被广泛应用于船只检测系统中。但是,红外成像系统容易受到烟幕干扰,导致船只检测系统失效。因此及时有效地检测到红外船只图像中的烟幕干扰区域,对于提高船只目标检测系统的准确性具有重要意义。针对红外图像中船只释放的烟幕区域的检测问题,提出了一种基于显著性与稠密光流融合的烟幕检测方法。由于舰船释放的烟幕有明显区别于背景的特性,因此首先采用多尺度邻域滤波的AC算法对图像进行显著性区域检测,提取显著的烟幕区域;然后利用烟幕扩散的运动特点,对图像前后帧进行稠密光流计算得到图像的运动信息,通过设置阈值筛选得到明显的运动点、扩充运动点区域、合并分裂的运动区域,得到运动的烟幕区域;最后对显著性区域与运动烟幕区域进行合并得到最终的烟幕区域。实验结果表明,该方法能有效检测到烟幕区域,并且能够很好地适应烟幕反射光以及背景亮度的变化。

Ship targets are important objects for marine monitoring,and infrared imaging system has been widely used in ship inspection systems due to its feature of working at the same time during the day and night.However,infrared imaging systems will be easily affected by the release of smoke screens,which result in the invalidation of ship detection systems.Therefore,timely and effective detection of the smoke interference area in the infrared ship image is of great significance for accurate ship target detection.Aiming at the problem of detecting the interference of smoke area from ships in infrared images,a smoke detection method based on the fusion of saliency and dense optical flow was proposed in the paper.Because the smoke screen released by the ship was obviously different from the background,the AC algorithm of multi-scale neighborhood filtering was firstly used to detect the saliency area of the image,and the significant smoke screen area was extracted.Then,the movement characteristics of the smoke screen were used to compare the front and back frame of the image sequences,and the frame dense optical flow was calculated to obtain the motion information of the image.By setting the threshold to filter the obvious motion points,expand the motion point area,merge the split motion areas,the motion smoke area was obtained.Finally,the saliency area and the motion area were fused,and the final smoke screen area was obtained.The experimental results show that the method can effectively detect the smoke screen area,and is able to adapt to the changes of the reflected light of smoke screens and the background brightness variations.