基于背景差分和混合帧差的运动目标检测

针对利用核密度估计建立背景模型时计算量大,运动目标和外界环境容易发生变化,提出一种基于改进的核密度估计背景差分法和改进的混合帧差法相结合的运动目标检测方法。该方法在背景建模时,先对背景差分后的图像进行分块和分类,并简化了核密度估计的核函数,对前景块中的像素进行核密度估计,减少了计算量。在混合帧差法中增加了动态阈值,提高了对光线变化的适应性。实验结果表明该方法能够完整地提取出运动目标,提高了目标检测的准确率。

Moving Object Detection Based on Background Subtraction and Hybrid Frame Differencing

Aiming at large amount of calculation in establishing background model based on kernel density estimation, and easy changes of moving object and external environment, this paper proposes a method of combining background subtraction based on improved kernel density estimation with improved hybrid frame difference for moving object detection. In the background modeling, this method first blocks and classifies the image through background subtraction, predigests the kernel function, and then performs kernel density estimation for the pixels belong to foreground block, reducing some computation. It adds dynamic threshold in the hybrid frame difference, and improves the adaptation to light. The experimental results show that the proposed algorithm can completely extract the moving objects, and improve the accuracy of object detection.