结合多尺度影像空间的帧差法运动目标检测

Moving Object Detection Method Combining Frame Difference with Multi-scale Image Space

针对帧差法检测运动目标不完整的问题，提出一种结合多尺度影像空间的帧间差分运动目标检测方法。该方法首先利用一组参数递增的高斯滤波器来获取三层多尺度影像，对不同尺度空间的影像采用Brox光流进行配准，进而对配准后的每一层影像做差值运算，将差值影像融合，以增强运动目标信息，并对增强后的差值影像进行阈值分割，最后采用形态学滤波去除少量噪声。实验结果表明：Brox光流能取得较好的运动补偿效果，且结合多尺度影像空间的运动目标检测方法能够较好地克服传统帧差法检测运动目标不完整的缺点。

A moving object detection method combining multi-scale image space with inter-frame difference is proposed to solve the problem of incomplete detection of moving objects using frame difference method. Firstly, a set of Gaussian filters with increasing parameters are used to acquire three-layer multl-scale image, and the optical flow of Brox is used to register the images in different scale. Secondly, the difference of each layer of image are calculated after registration, and difference images are fused to enhance the information of moving objects. Besides, enhanced images are processed using threshold segmentation method. Fi- nally, morphological filtering is used to remove some noise. The experiment results show that the optical flow of Brox can provide better motion compensation, and the moving object detection method combined with multi-scale image space can overcome the shortcoming of incomplete detection of moving objects in traditional frame difference method.