摘要
针对目标模型内的背景像素造成目标跟踪定位偏差的问题,提出了一种适合于目标跟踪的模糊核直方图。通过在核直方图中引入模糊隶属度,可降低背景像素对目标特征的影响。另外,研究了模糊隶属度函数,根据中心-周围方法提取背景特征,并给出了确定模糊隶属度函数的两种策略,即比率策略和差分策略。在比率策略中采用对数似然函数技术,而在差分策略中采用目标特征与背景特征的差分技术。实验表明,基于比率策略的目标跟踪更适合于场景简单的情形,而基于差分策略的目标跟踪适合于场景复杂的情形。
To resolve the problem that the background pixels in an object model induce localization errors in object tracking, a fuzzy kernel histogram was presented for object tracking. The fuzzy membership degree was introduced in the fuzzy kernel histogram for reducing the localization errors in object tracking produced by background pixels. Moreover, the fuzzy membership functions were studied, and two strategies, the ratio strategy and difference strategy, were given for determining the fuzzy membership functions after extracting the background feature using a ＂center-surround＂ approach. The log-like- lihood function was adopted in ratio strategy, while a difference technique was introduced in difference strategy. The experimental results show that the object tracking based on the ratio strategy is fit for the case of simple scene, while the object tracking based on the difference strategy is fit for the case of complex scene.
出处
《高技术通讯》
CAS
CSCD
北大核心
2009年第2期174-180,共7页
Chinese High Technology Letters
基金
国家自然科学基金(60234030,60773110)
国家基础研究(A1420060159)资助项目
