Current Issue Cover
区域颜色属性空间直方图背景建模

金静,党建武,王阳萍,翟凤文(兰州交通大学)

摘 要
摘 要:目的 为了能在光照变化、动态背景干扰这一类复杂场景中实时、准确地分割出运动前景,针对传统的基于颜色特征和基于像素的方法的不足,提出一种在颜色属性空间进行区域直方图建模的运动目标检测方法。方法 首先将RGB颜色空间映射到更为稳健的低维颜色属性空间,以颜色属性为特征在像素的局部范围内建立直方图,同时记录直方图每一个分区中像素的空间信息,使用 个空间直方图构成每个像素的背景模型,每个直方图根据其匹配度赋予不同的权重。降维的颜色属性提高了模型的鲁棒性和检测的时效性,空间直方图引入的位置信息提高了背景模型的准确性。然后通过学习率 和 来控制各模型直方图及其权重的更新,以提高模型的适应性。在标准测试数据集的所有视频序列中进行了实验,通过分析综合性能指标(F1)及平均假阳性(FN)曲线,确定了算法中涉及参数的合理取值范围。结果 对实验结果定性和定量的分析表明,方法能够得到良好的前景检测效果,尤其在多模态场景和光线变化的复杂场景中能显著提高检测性能。各类场景的平均综合性能指标(Average F1)相比性能突出的方法ViBe、LOBSTER和DECOLOR分别提高了0.65%、3.86%和3.9%,并通过GPU并行加速能够实现运动目标的实时检测。结论 在复杂视频环境下的运动目标检测中,相比已有方法,本文方法能够更为准确地分割出运动前景,是一种实时的、有效的检测方法,具有一定的实用价值。
关键词
Region Spatiogram in Color Names for Background Modeling

Jin Jing,Dang Jianwu,Wang Yangping,Zhai Fenwen(School of Electronic and Information Engineering,Lanzhou Jiao Tong University,Lanzhou)

Abstract
Abstract: Objective In recent years, the technique of Intelligent Video Analysis is a very important research area in Com-puter Vision. Moving objects detection aims to catch moving foreground in all kinds of surveillance environment, which is an essential foundation for following video processing such as targets tracking, objects segmentation and so on. The tradi-tional methods often model the background in color feature space and single pixel. The traditional color feature would be disturbed by light and shadow easily. The single pixel cannot reflect the region spatial relation between pixels. In order to detect the moving foreground precisely in complex video sequences including illumination and dynamic background in time, the paper proposed a moving detection method based on background modeling technique by region spatiogram in Color Names space. Color Names are linguistic labels that humans attach to colors. The learning of the Color Names is achieved by the PLSA model. In fact, it conducts a mapping from RGB space to robust 11 dimension CN space. Modeling back-ground in Color Names space will deal with the illumination variation better. Histogram is a zeroth-order tool for feature description that is robust to scale variation and rotation variation, while second-order spatiogram contains spatial mean and covariance for each histogram bin. So the spatiogram retains more information about the geometry of the patches and cap-tures the global positions of the pixels rather than their pairwise relationships. Therefore, it is necessary to employ spatio-gram in Color Names space for background modeling. Method In this paper, a novel method was proposed for moving detection. At first, it mapped RGB color space to lower-dimensional Color Names space that is more robust. Then it estab-lished spatiogram in pixel local region characterized by Color Names feature and recorded spatial information of pixels in every bin. The background models of every pixel are constituted by spatiograms. Every spatiogram was given different weight according to its matching rate. Color Names feature by dimension reduction enhanced the robustness of models and detecting timeliness. Spatial information introduced by spatiogram enhanced the accuracy of the background model. To en-hance adaptivity of models, the approach controlled update of model spatiogram and their weights by learning rate and . Then it performed experiments at all video sequences from standard test data CDnet that included different chal-lenges such as illumination variation, moving shadow, multi-model background and so on. The parameters such as model size , threshold , and learning rate , in the algorithm were determined by analysis from comprehen-sive performance F1 and averaged FN (False Negative) curves. Result The quantitative and qualitative analysis to experi-ments indicates that the method can achieve expected results. It can get outstanding effect in some scenes including illumi-nation and multi-model background especially. Comparing to ViBe、LOBSTER and DECOLOR , the method enhances 0.65%, 3.86% and 3.9% of average comprehensive performance F1 of all scenes respectively. Modeling for every pixel in its local region is concurrent. So real-time detection is achieved with GPU parallel acceleration in order to improve time ef-ficiency. Conclusion Robust Color Names space cope with illumination variation better. Multiple spatiogram models match multi-model background better such as waving tree, water and fountain. Therefore, the algorithm can segment the moving foreground more accurately in complex video environment compared with the existing methods. It is a real-time and effec-tive detection algorithm .It has certain practical value in Intelligent Video Analysis.
Keywords
QQ在线


订阅号|日报