[1]逯鹏,张利亚,黄石磊,等.基于稀疏超完备的异常行为检测算法[J].郑州大学学报(工学版),2016,37(06):72-76.[doi:10.13705/j.issn.1671-6833.2016.03.031]
 Lu Peng,Zhang Lia,Huang Shilei,et al.Abnormal Behavior Detection Algorithm Based on Sparse Overcomplete Representation[J].Journal of Zhengzhou University (Engineering Science),2016,37(06):72-76.[doi:10.13705/j.issn.1671-6833.2016.03.031]
点击复制

基于稀疏超完备的异常行为检测算法()
分享到:

《郑州大学学报(工学版)》[ISSN:1671-6833/CN:41-1339/T]

卷:
37卷
期数:
2016年06期
页码:
72-76
栏目:
出版日期:
2016-11-30

文章信息/Info

Title:
Abnormal Behavior Detection Algorithm Based on Sparse Overcomplete Representation
作者:
逯鹏张利亚黄石磊李奇航张微
郑州大学 电气工程学院,河南 郑州,450001
Author(s):
Lu Peng Zhang Lia Huang Shilei Li Qihang Zhang Wei
School of Electrical Engineering, Zhengzhou University, Zhengzhou, Henan 450001
关键词:
Keywords:
DOI:
10.13705/j.issn.1671-6833.2016.03.031
文献标志码:
A
摘要:
视频异常行为检测的核心问题是如何从海量数据中理解行为。提出一种新的基于视觉稀疏超完备表示机制的特定场景中视频异常行为检测算法,针对视频中感兴趣的点提取局部有效信息,提高数据处理效率。首先,提取训练样本的时空兴趣点,计算局部时空特征;其次,将该特征输入稀疏超完备模型,训练得到一组稀疏基;然后,利用上述基对待测视频进行重构,通过局部时空特征重构的差异实现异常行为检测;最后,提出对不同视频的稀疏基更新算法。标准数据库的实验表明,该算法能够有效解决异常行为检测问题,检测准确率高,错误警报率低。
Abstract:
The focus question of video abnormal behavior detection was how to illustrate the behavior correctly through analysis of huge amounts of data. A new algorithm was proposed based on visual sparse overcomplete representation mechanism to extract local effective information about the interest points in the video of specific scenario, which could improve the efficiency of data processing. Firstly, the algorithm extracted the local spa-tial temporal interesting points ( STIP) in training samples. At the same time it calculated the local spatial temporal characteristics. Then it put the characteristics into sparse overcomplete representation model to get a set of sparse matrix after training. Finally, it reconstructed the query video using the aforementioned matrix to detect abnormal behavior through the reconstruction error of local spatial temporal characteristics. In addition, the updated algorithm of sparse matrix function for different videos was proposed. Experiment results on stand-ard database showed that our algorithm could detect abnormal behavior effectively and with higher accuracy and lower false alarm.
更新日期/Last Update: