基于低秩逼近的视频序列中的异常事件检测

发布时间：2018-04-23 22:14

本文选题：低秩逼近 + 视频分析　；参考：《南京理工大学》2017年硕士论文

【摘要】：近年来,低秩逼近理论在模式识别和计算机视觉领域已得到广泛应用,比如:协同过滤,图像配准、视频去噪等研究,但是很少有研究将其用于视频序列中的异常事件检测。本文在低秩逼近算法理论的基础上,针对视频数据的低秩特性,将低秩逼近算法改进应用于视频序列中的异常事件检测当中。本文的主要工作和研究内容如下:(1)提出了基于低秩结构化稀疏表示的异常事件检测方法。如何从庞大的视频数据中挖掘行为模式的关键信息而摒弃冗余内容,以及提高异常事件的检测效率是本文所要解决的关键问题。针对该问题,本文引入了低秩组稀疏编码(LASSC)模型,提出了一种基于低秩结构化稀疏表示的异常事件检测方法。实验结果表明,该方法提高异常事件检测准确率,并且还极大地改善时间效率。(2)提出了基于低秩结构化稀疏表示的自适应异常事件检测方法。基于低秩结构化稀疏表示的异常事件检测方法有时候会将正常测试样本检测为异常事件,导致出现异常事件检测的错误接收率较高的情况。考虑到字典学习过程中的低秩信息没有被充分利用,本文实现了将低秩组稀疏字典和对应的低秩信息一起用于视频事件重建,提出了一种基于低秩结构化稀疏表示的自适应异常事件检测方法。该方法包括两个主要阶段:基于低秩结构化稀疏表示的字典学习以及基于低秩组稀疏字典的加权视频事件重建。实验结果表明,该方法在保证较高时间效率的同时还有效地提高了检测准确率。(3)实现了一种改进的基于低秩结构化稀疏表示的自适应异常事件检测方法。针对直接利用LASSC模型进行字典学习导致时间效率不高的问题,本文把低秩组稀疏字典作为初始字典进一步拟合目标函数,提出一种改进的基于低秩结构化稀疏表示的自适应异常事件检测方法。实验结果表明,该方法在保证检测准确率的同时还提高了时间效率。
[Abstract]:In recent years, low rank approximation theory has been widely used in the fields of pattern recognition and computer vision, such as collaborative filtering, image registration, video denoising, etc. In this paper, based on the theory of low rank approximation algorithm and considering the low rank characteristic of video data, the low rank approximation algorithm is improved to detect abnormal events in video sequences. The main work and research contents of this paper are as follows: 1) an anomaly event detection method based on low rank structured sparse representation is proposed. How to mine the key information of behavior pattern from the huge video data and abandon redundant content and improve the detection efficiency of abnormal events are the key problems to be solved in this paper. In order to solve this problem, this paper introduces the low rank group sparse coding (LASSC) model, and proposes an anomaly event detection method based on low rank structured sparse representation. Experimental results show that the proposed method improves the accuracy of anomaly detection and greatly improves the time efficiency.) an adaptive anomaly event detection method based on low rank structured sparse representation is proposed. The method of anomaly event detection based on low rank structured sparse representation sometimes detects normal test samples as abnormal events, which leads to high error reception rate of abnormal event detection. Considering that the low rank information in the dictionary learning process is not fully utilized, this paper proposes a low rank sparse dictionary with the corresponding low rank information for video event reconstruction. An adaptive anomaly event detection method based on low rank structured sparse representation is proposed. The method includes two main stages: dictionary learning based on low rank structured sparse representation and weighted video event reconstruction based on low rank sparse dictionary. Experimental results show that the proposed method not only ensures high time efficiency but also improves the detection accuracy effectively. It implements an improved adaptive anomaly event detection method based on low rank structured sparse representation. In order to solve the problem of low efficiency due to the direct use of LASSC model in dictionary learning, this paper takes the low rank sparse dictionary as the initial dictionary to further fit the objective function. An improved adaptive anomaly event detection method based on low rank structured sparse representation is proposed. The experimental results show that the method not only ensures the detection accuracy but also improves the time efficiency.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】