基于目标存在概率场的多视角运动目标检测与对应算法  

Object Occupancy Probabilistic Field Based Multi-view Moving Object Detection and Correspondence

作　　者：徐剑[1] 丁晓青[1] 王生进[1] 

机构地区：[1]清华大学电子工程系智能技术与系统国家重点实验室,北京100084

出　　处：《自动化学报》2008年第5期609-612,共4页Acta Automatica Sinica

基　　金：国家自然科学基金(60472002)资助~~

摘　　要：提出了一种多视角运动目标检测与对应算法—基于目标存在概率场的运动目标检测和对应算法.目标存在概率场是算法的核心,其融合各视角的信息,并依据融合结果在空间中检测目标.目标窗结构是快速计算目标存在概率场的关键,并给出了一种同时实现目标检测和对应的框架.实验结果表明该算法具有遮挡处理能力强,计算复杂度低的优点.A novel algorithm to detect objects across multiple views at the same instance is proposed based on object probabilistic random field （OOPF）. OOPF is the core of our algorithm and it integrates all the information of individual views into the fusion space. Then object detection is implemented in the fusion space. An object window structure is designed in order to compute OOPF fast, which also provides a scheme to detect objects and the correspondence simultaneously. The experiment results demonstrate that the algorithm has a better performance against occlusion while the computational complexity is low.

关 键 词：目标检测 目标对应 多视角 存在概率 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]