鲁棒主成分分析的运动目标检测综述

发布时间：2018-01-08 16:24

本文关键词：鲁棒主成分分析的运动目标检测综述　出处：《中国图象图形学报》2016年10期 　论文类型：期刊论文

【摘要】：目的运动目标检测在许多计算机视觉任务中发挥了重要的作用。背景建模是运动目标检测中传统而又常用的方法。然而,许多背景建模方法是基于像素点的,对背景方面的考虑过于简单,难于处理真实视频。最近,将基于低秩和稀疏分解的鲁棒主成分分析应用于运动目标检测成为计算机视觉领域内的研究热点。为使更多国内外运动目标检测的研究者对鲁棒主成分分析方法进行探索和应用,本文对其进行系统综述。方法融入最新研究进展,基于误差抑制、贝叶斯理论、时间和空间信息、多特征和多因素耦合,对各种国内外的鲁棒主成分分析模型进行归纳,并理论分析其优缺点。结果本文采用变化检测数据集(change detection dataset)中不同场景的视频序列来对不同算法进行对比实验。从实验结果可知,属于第3类方法的DECOLOR的检测效果优于其他算法,在均值对比中得到的召回率、精确率和F-measure分别为0.7、0.706和0.66。总体来说,当前改进算法都能有效地弥补最初鲁棒主成分分析方法的缺陷,提高了运动目标检测的精度。结论鲁棒主成分分析在运动目标检测上取得了较多的研究与应用成果,在智能视频监控应用领域拥有广阔的应用前景。但是,其仍需针对鲁棒主成分分析存在的一些局限性进行深入的研究。融入前景运动目标在视频中的先验知识是基于鲁棒主成分分析的运动目标检测的发展趋势。
[Abstract]:The purpose of moving target detection plays an important role in many computer vision tasks. Background modeling is the traditional method and moving target detection in common. However, many of the background modeling method is based on the pixels, the background considerations are too simple, it is difficult to deal with real video. Recently, the robust principal component and sparse low rank decomposition analysis is applied to detect moving objects has become a research hotspot in the field of computer vision. It is based on a more domestic and foreign researchers are moving target detection exploration and application of analysis method of robust principal component, this paper systematically reviewed the methods of integration. The latest research progress, based on the error suppression, Bayesian theory, time and space the information, coupled with multi feature and multi factors, robust principal component analysis model of domestic and abroad are summarized, and the theoretical analysis of its advantages and disadvantages. The results of this paper. With the change of detection data set (change detection dataset) video sequence in different scenes to the comparison of the different algorithms. The experiment results showed that the detection effect is better than the DECOLOR method belongs to the third kind of other algorithms in the mean contrast obtained in the recall rate, precision rate and F-measure were 0.7,0.706 and 0.66. in general, current the improved algorithm can effectively compensate for the defects of original robust principal component analysis method, improves the accuracy of motion detection. Conclusion robust principal component analysis has made many research achievements and applications in moving target detection, have broad application prospects in intelligent video surveillance applications. However, it is still needed for robust principal component some analysis of the limitation of the in-depth study. In the video moving target with prior knowledge is robust principal component analysis based on the motion The development trend of target detection.

【作者单位】：广东工业大学信息工程学院;
【基金】：国家自然科学基金项目(61001179,61571139) 广东省科技重大专项基金项目(2015B010124001) 广东省工程技术研究中心项目(2015B090903017) 广东省自然科学基金项目(2015A030312008)~~
【分类号】：TP391.41
【正文快照】： Vol.21,No.10,Oct.20160引言运动目标检测[1-2]是智能视频分析中关键的第1步,对许多高层次的计算机视觉任务起着基础性作用,如目标跟踪[3]、行为识别[4]、场景分析[5]、交通监控[6]等。运动目标检测的基本任务是将前景运动目标从观测视频图像中提取出来,也就是说,将前景和背景

【相似文献】