基于改进C-V模型的多运动目标分割

Segmentation of multiple moving targets based on modified C-V model

针对固定摄像机视频监控中多运动目标自动分割问题,本文提出了一种基于帧间差分和改进C-V模型的新方法。首先,通过自适应阈值获得滤波后的相邻帧差值图像。其次,通过连通域分析和设定阈值,消除差值图像中噪声的影响并标定出运动目标所在的区域,计算运动区域的运动窗口。最后,对所有运动窗口,采用改进的C-V模型作分割,得到封闭和完整的运动目标轮廓。针对C-V活动轮廓模型不能自适应地分割非匀质图像问题,提出利用全局梯度信息演化活动轮廓曲线,根据闭合活动轮廓曲线内外部的梯度信息重新定义图像分割能量函数。实验结果表明,该算法避免了对整个图像的分割,减少了运算量,能实现对刚体或非刚体的多运动目标的自动检测和轮廓分割。

Based on inner-frame difference and modified C-V model, a novel segmentation method is presented for multiple moving targets in static camera surveillance. Firstly, the adaptive threshold is used to get the difference image between two adjacent filtered frames. Then by analyzing the connected regions and setting a threshold, small background noise regions are removed, moving target regions are marked and motion-changed regions are determined from the moving target regions. Finally, the modified C-V model is adopted to segment each motion-changed region, and the closed and entire target contour is obtained. In order to solve the problem that the C-V active contour model can not adaptively segment non-homogenous image, the global gradient information is employed to evolve the active contour curve, and a new energy function is defined according to the gradient information inside and outside the closed active contour curve. Experiments show that this approach avoids segmenting the whole image, reduces computation time and can realize automatic detection and contour segmentation of rigid or non-rigid moving targets in static camera surveillance.