波纹干扰抑制下内河CCTV系统运动船舶检测

Ripple-Driven Ship Detection in Inland Waterway CCTV System

传统运动船舶检测方法不能较好的描述水面波纹起伏运动背景问题。针对水面波纹扰动问题,提出了一种融合视觉注意机制和改进的混合高斯背景建模技术的综合船舶检测策略。利用运动船舶导致水域空间不连续的特点,根据视觉注意机制生成显著图。船舶目标敏感区域对应较高的显著度,空域连续的水面区域呈现较低的显著度,对合成显著图进行自动阈值分割滤除波纹背景区域。将视觉注意机制与基于改进的混合高斯建模背景减除法检测结果进行融合得到运动船舶在当前帧的位置。实验结果表明,改进算法的船舶检测性能明显优于传统的混合高斯背景建模方法,同时对水面波纹扰动的鲁棒性能也显著提高。

Traditional background model methods can not model the ripples well. A hybrid saliency -based visual attention method and mixture of Gauss background model is presented to tackle the inevitable ripples. Saliency figures are obtained by the continuous motion between consecutive frames. Because the moving ship of interest gets higher saliency and the background gets lower, thus we employe a threshold manner to separate the foreground and back- ground. Then we fuse the detection results of saliency - based visual attention and mixture of Gauss background mod- el. The experimental results demonstrate that the proposed method achieves comparatively higher accuracy than origi- nal background subtraction method. Simultaneously, our method is more robust to ripples.