在复杂场景中,阴影严重影响目标检测的准确度,传统目标检测方法易将运动阴影误检为运动目标。针对这一问题,提出了改进的拉普拉斯-高斯(log)算子与ViBe算法相结合的视频阴影消除算法。在log算子检测的基础上,采用二维数字滤波器进行滤波检测;将二维数字滤波器的系数矩阵旋转180°以创建卷积核;利用二维卷积对创建的卷积核与输入的图像矩阵进行卷积,实现对图像的边缘检测,最后利用改进的log算子提取外轮廓边缘和前景目标边缘,二者相减得到内部边缘,即消除阴影的运动目标边缘。实验结果表明,改进的算法可有效地消除阴影,同时提高了算法的鲁棒性。

Shadows seriously affect the target detection accuracy in a complex scene. A traditional target-detection method can easily misidentify moving shadows as moving targets. To solve this problem, this study proposes an improved video-shading removal algorithm combining the Laplace-Gaussian (log) operator with the ViBe algorithm. A two-dimensional (2D) digital filter based on the log operator was used for filter detection. The coefficient matrix of the 2D digital filter was rotated by 180° to create a convolution kernel, which was convoluted with the input image matrix to achieve image edge detection. Finally, the improved log operator was used to extract the outer contour and foreground target edges. The two edges were subtracted to obtain the inner edge, which was the moving target edge with the shadow eliminated. Experiment results reveal that the improved algorithm is robust and can effectively eliminate shadows.