摘要
真实场景的视频目标检测需要消除阴影、反射和鬼影等噪声的影响,以检测出运动目标和静止目标.为了实现系统性的视频目标检测,提出一种自适应圆锥裁剪联通块(TC-BLOB)检测方法.基于BLOB知识,将3D颜色空间变换为"夹角-模差"2D空间后,定义一套圆锥裁剪规则划分出阴影BLOB和反射BLOB;再以一种持久化记忆PM方法判别出鬼影BLOB;最后改进双背景模型检测出静止目标和运动目标.采用不同环境视频进行实验的结果表明,文中方法是有效的,并有独立于后续跟踪的优点.
Segmentation of objects in real-world scenes needs to detect and eliminate noise such as shadow, reflection and ghost. The detection of static visual object is also crucial for robust segmentation and dynamic background updating. To realize systemic object segmentation for real-world scenes, the paper proposed a new adaptive method based on truncated cone BLOB. Firstly, based on BLOB knowledge, the method transferred 3D color space to 2D angle mode color space and defined a set of rules of truncated cone to detect shadow BLOB and reflection BLOB. Secondly, the method removed the ghost BLOB from the moving visual BLOB by calculating permanence memory. Finally, the method used a long-time and short time dual-background model to detect static visual object. The experimental results on different scenes demonstrate the effectiveness of the proposed method. In addition, the proposed method can be used for different tracking techniques.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2009年第9期1344-1351,共8页
Journal of Computer-Aided Design & Computer Graphics
基金
国家"八六三"高技术研究发展计划(2006AA12A104)
关键词
TC—BLOB目标检测
持久化记忆
鬼影检测
静止目标检测
object detection based on truncated cone BLOB
permanence memory
ghost detection
static visual object detection
