基于朗斯基函数的混合高斯模型运动目标检测*

针对传统的混合高斯模型在进行运动目标检测时存在拖影和性能差的缺点,本文提出了一种融合朗斯基函数和帧间差分法的混合高斯背景建模算法。该改进算法通过朗斯基矩阵行列式判断相邻像素间空间域相关性,以此增加模型参数更新条件,改进模型参数更新机制；并利用帧间差分法检测运动目标轮廓的灵敏性,将两种检测结果布尔或运算,完善目标轮廓。实验结果表明,该改进算法对拖影现象达到很好的抑制作用,并使算法检测性能得到提高。

New method for mixture gaussian background model and moving object detection based on Wronskian function

Considering the shortcomings, e.g., trailing smear and lower performance, of the traditional GMM in the process of moving object detection, this paper proposed an improved Gaussian mixture model algorithm fusing Wronskian function and frames difference method. After the process of Gaussian mixture model, it judged the spatial domain correlation between neighboring pixels by the value of the Wronskian matrix determinant which increased the update condition of the model parameters and improved the updating mechanism of the model parameters. Finally, using the sensitivity of frames difference method to detect moving target contour, it applied Boolean OR operation on the results of improved GMM and frames difference method to get the preferable moving object. The experimental results show that the improved algorithm can effectively suppress the phenomenon of the trailing smear and enhance the detection performance.