摘要
针对视频分割中的阴影消除问题,提出了一种以置信度为桥梁,前景边缘投影特征与局部纹理特性相融合的阴影提取算法.采用自适应高斯法获得动态背景,提取包含阴影的前景,计算出当前帧和背景帧在前景最小外接矩形坐标范围内的边缘差异,得到低干扰的车辆和阴影边缘信息.利用大津阈值算法进行投影分割,在阴影连续性前提下,高置信度区域确认为阴影,低置信度区域确认为车辆,而一般置信度区域,进一步结合局部纹理在当前帧和背景帧间的跳变程度,搜索出与车辆相连的阴影.结果表明:该方法能够去除导致前景严重变形的大面积阴影,去除有效率在90%以上,保障了车辆的有效提取;算法实时性好,可应用于智能视频监控的目标检测及跟踪中.
To achieve shadow elimination in video target segmentation, according to confidence, charac- ters of foreground edge projection and local texture features were merged to propose a novel shadow ex- traction strategy. By adaptive Gaussian method, the foreground containing shadow was obtained to extract dynamic background. The edge difference between foreground and background in minimum enclosing rec- tangle area of foreground was calculated to achieve edges and shadow of vehicle. Otsu algorithm was used for video segmentation. The area with high confidence was labeled as shadow, and that with low confi- dence was labeled as vehicle. According to the jumping level of local texture between current frame and background frame, the remaining shadow concerning with vehicle was found out in the area with middle confidence by further processing. Experimental results demonstrate that the proposed method can effec- tively remove huge shadows which may lead to heavy deformation with over 90% elimination rate. The al- gorithm can be applied in object detecting and tacking in intelligent video surveillance system with good real-time performance.
出处
《江苏大学学报（自然科学版）》
EI
CAS
北大核心
2012年第2期144-149,共6页
Journal of Jiangsu University：Natural Science Edition
基金
国家科技支撑计划项目(2009BAG13A04)
江苏省自然科学基金资助项目(BK2010239)
关键词
智能交通系统
阴影消除
车辆跟踪
特征融合
纹理
高斯分布
intelligent transportation system
shadow elimination
vehicle tracking
features integrating
textures
Gaussian distribution
