摘要
为了能够从视频序列中快速准确地检测运动目标,在混合高斯背景差分法的基础上引入Grabcut算法,提出了一种新的运动目标检测G-GMM(Grabcut-Gaussian Mixture Model)算法。首先通过混合高斯模型背景差分法提取运动目标初始二值轮廓,构建其最小的外接矩形;然后初始化矩形内图像信息,寻找潜在前景区域;最后采用迭代算法实现最优化分割,得到准确的运动目标轮廓。实验结果表明,在静止摄像机户外视频监控系统中,提出算法具有较高的准确性和鲁棒性,对刚性和非刚性两类目标都具有较好的检测结果。
To detect moving objects accurately and rapidly from the videos sequences, this paper proposed a novel G- GMM method for automatic detection via combination of GMM and Grabcut techniques in image processing. Firstly, this algorithm uses GMM（Gaussian Mixture Model） based background subtraction to produce binary images for every moving object and then constructs their minimum marking rectangles. And then it follows the image information initialization of each marking rectangle via Grabcut. Finally, an iterative algorithm with foreground parameters is adopted to optimize the object segmentation and thus the moving object contour is obtained. Experimental results indicate that the proposed method achieves good accuracy and robustness in the still camera outdoor video surveillance system, providing promising detection results for both rigid and non-rigid objects.
出处
《计算机科学》
CSCD
北大核心
2015年第B11期199-202,共4页
Computer Science
基金
国家自然科学基金(61502331)
天津市应用基础与前沿技术研究计划(15JCQNJC00800)
天津财经大学优秀青年学者计划项目
天津财经大学"大学生创新创业训练计划"项目(201410070014)资助
关键词
目标检测
GGMM
运动前景
图像分割
Object detection, Grabcut-gaussian mixture model（G-GMM）, Foreground motion, Image segmentation
