为了在复杂天空背景下检测出低空慢速小目标, 本文研究了“低小慢”目标的视觉显著性区域特征, 融合扫描线填充算法, 提出了一种动态背景下“低小慢”目标自适应实时检测技术。首先, 根据图像的亮度对比度获取显著性图。接着, 使用形态学梯度提取显著性特征, 通过三帧差分算法得到种子点。然后, 使用扫描线填充算法进行生长, 结合提出的自适应双高斯算法分割出前景。最后, 根据候选目标的面积占比变化、质心距离变化、宽高比差异剔除虚假目标, 完成检测。为了验证算法的有效性, 本文选取了7组复杂天空背景的视频序列进行测试, 并与其他优秀检测算法进行了对比。结果表明, 本文提出的算法对运动目标检测的平均运行时间为0040 9 s, 平均检测准确率为8997%, 相比于其他算法的平均运算时间减少了035 s, 检测的平均准确率提高了245%。算法在复杂背景下具有较好的稳定性和较强的鲁棒性。

In order to detect LSS(Low, Small and Slow) targets in complex sky backgrounds, we study the visual salient region characteristics of the LSS target and scan line filling algorithm and propose an adaptive real-time detection technology for LSS targets in dynamic complex backgrounds. Firstly, a saliency map is obtained based on the Luminance Contrast(LC) of the image. Secondly, the morphological gradient is used to extract the saliency feature and the seed points of the scan line filling algorithm are obtained by the three frame difference algorithm. Then, the scan line filling algorithm is used to grow the image and the foreground is segmented using the proposed adaptive double Gauss threshold segmentation algorithm. Finally, according to the change of the object′s area of occupation, the center distance and the aspect ratio of the candidate target, the false targets are eliminated and detection is completed. In order to verify the effectiveness of the algorithm, 7 test groups of complex sky background video sequences are selected and compared with other excellent detection algorithms. The results show that the running time of the proposed algorithm for moving object detection is 0040 9 s and the accuracy rate is 8997%. When compared with other algorithms, the average running time is reduced by 035 s, and the average accuracy of detection is enhanced by 245%. The algorithm has good stability and is robust in target detection in complex backgrounds.