基于块模型的混合高斯运动目标检测方法

A Gaussian Mixture Moving Object Detection Method Based on Block Model

对场景中运动目标进行实时检测,在公共安全、军事和航空航天等领域中具有非常重要的应用价值。在图像处理运动检测中,传统的混合高斯模型方法为每一个像素构建高斯模型,利用背景像素在较长时间内具有稳定的概率分布密度等统计信息分离背景区域,实现背景和运动目标的分离。该方法在实际使用中容易受到噪声干扰,且计算成本高。针对其不可避免的缺点,提出了一种基于信息度对图像进行分块的混合高斯模型算法。通过图像块中含有运动点的比例不同采用不同的混合高斯模型,对位于运动目标边缘的图像块采用单独的算法进行预处理,不仅能够降低背景区域、运动区域中噪声对背景模型的影响,同时能够以更高的计算效率实现运动目标的检测。实验结果表明,该算法在实际场景中具有良好的可行性和鲁棒性。

Real-time detection of moving targets in the scene is of great application value in the fields of public security,military and aerospace.In the motion detection of image processing,the traditional mixed Gaussian model constructs a Gaussian model for each pixel,and uses the background pixel with stable probability distribution density and other statistical information for a long period to separate the background region,so as to achieve the separation of background and moving targets.This method is subject to be interfered by noise in practical use and has high calculation cost.For this,we propose a mixed Gaussian model algorithm for image segmentation based on information degree.Different mixed Gaussian model are adopted according to different proportions of moving points in the image blocks,and the image blocks located at the edge of moving target are preprocessed by a separate algorithm.It can not only reduce the influence of noise in the background area and moving area on the background model,but also realize the detection of moving targets with higher computational efficiency.The experiment shows that the proposed algorithm is feasible and robust in practical scenario.