摘要
为了提高运动目标检测的准确性和完整性,在背景建模研究的基础上,提出一种结合非参数随机样本集ViBe算法和核图割KGC(Kernel graph cuts)理论的运动目标检测方法。采用改进的ViBe算法进行背景提取,把样本集扩展到阴影空间,抑制目标运动阴影的产生。针对样本随机性和噪声带来的目标不完整性,建立目标边缘轮廓分割约束,对前景区进行KGC分析,把滑动窗局限于目标轮廓最大概率密度区,过滤无效背景,合并有效区域。经过视频测试实验表明,该方法在不增加较大计算的情况下,达到检测实时运行,相对于传统算法获得较好的准确率、完整性和鲁棒性。
To improve the accuracy and completeness of moving object detection , a KGC background subtraction algorithm combined with ViBe is proposed based on background modeling research .The background extraction is conducted under improved ViBe, which suppresses the generation of moving shadow through extended sample set for shadow space.The detected object is incomplete due to the effect of samples random and noise .The foreground region is processed with KGC by establishing object silhouette constraint of segmentation .The sliding window is confined in the maximum probability of object boundary as to filter invalid background and merge true regions .The experiments show that the proposed algorithm runs in real-time without aggravating computation load , and performs favorably against traditional algorithms on video sequences in terms of accuracy, completeness and robustness.
出处
《电子测量与仪器学报》
CSCD
2014年第7期695-702,共8页
Journal of Electronic Measurement and Instrumentation
基金
国家"863"计划(2012AA7041003)资助项目
关键词
运动目标检测
背景建模
核图割
阴影抑制
moving object detection
background modeling
kernel graph cuts
shadow suppression
