一种动态场景下基于时空信息的视频对象提取算法  

Video Object Extraction Algorithm Based on Spatio-temporal Information in Dynamic Scene

作　　者：田宏阳[1] 陈辉[1] 马文静[1] 

机构地区：[1]山东大学信息科学与工程学院,济南250100

出　　处：《中国图象图形学报》2007年第9期1653-1658,共6页Journal of Image and Graphics

基　　金：山东省优秀中青年科学家科研奖励基金(03BS001);山东省自然科学基金重点项目(Z2005G02)

摘　　要：在实际应用中,许多视频序列具有运动背景,使得从其中提取视频对象变得复杂,为此提出了一种基于运动估计和图形金字塔的动态场景下的视频对象提取算法。该算法首先引入了相位相关法求取运动向量,因避免了视频序列中光照变化的影响,故可提高效率和稳健性;接着再根据参数模型进行全局运动估计来得到最终运动模板;然后利用图形金字塔算法对当前模板内图像区域进行空间分割,最终提取出语义视频对象。与现有算法相比,对于从具有动态场景的视频流中提取运动对象的情况,由于使用该算法能有效地避开精准背景补偿,因而不仅节省了计算量,而且提取出来的语义对象精度较高。实验表明,无论是对动态场景中刚性还是非刚性运动物体的分割,该算法都具有较好的效果。In practical applications, many video sequences have moving background, and then the extraction of video object becomes complicated. An algorithm is proposed in the paper to extract video object from dynamic scene based on motion estimation and the graph pyramid. Phase correlation is first used to obtain the motion vector with high efficiency and robustness, and to weaken the impacts of illumination in the video sequence. Then global motion estimation with parameter-model is used to find the final motion template. Finally, to extract the semantic video object, spatial segmentation using the graph pyramid is applied to the image region in the current motion template. Compared with some prevailing methods, in the case of extraction of moving object from video sequences of dynamic scene, our algorithm avoids precise background compensation and is very computationally efficient, while the extracted semantic object is of high precision. The experimental results show that both rigid and non-rigid moving objects in dynamic scene are well extracted by this algorithm.

关 键 词：动态场景 视频对象提取 相位相关 运动估计 图形金字塔 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TN919.81[自动化与计算机技术—计算机科学与技术]