Real-time and accurate abnormal behavior detection in videos
- PDF / 2,415,172 Bytes
- 13 Pages / 595.276 x 790.866 pts Page_size
- 112 Downloads / 248 Views
ORIGINAL PAPER
Real-time and accurate abnormal behavior detection in videos Zheyi Fan1 · Jianyuan Yin1 · Yu Song1 · Zhiwen Liu1 Received: 15 December 2018 / Revised: 19 June 2020 / Accepted: 17 August 2020 / Published online: 24 September 2020 © Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract Abnormal crowd behavior detection is a hot research topic in the field of computer vision. In order to solve the problems of high computational cost and the imbalance between positive and negative samples, we propose an efficient algorithm that can detect and locate anomalies in videos. In order to solve the problem of less negative samples, the algorithm uses the spatiotemporal autoencoder to identify and extract the negative samples (contain abnormal behaviors) in the dataset in an unsupervised learning method. On this basis, a spatiotemporal convolutional neural network (CNN) is constructed with simple structure and low computational complexity. The supervised training method is used to train the spatiotemporal CNN with positive and negative samples to generate the detection model. Experiments are conducted on the UCSD and UMN datasets. The experiment results show that the proposed algorithm can detect and locate abnormal behaviors in real time (using only CPU), and the accuracy of the algorithm exceeds those of the existing algorithms at both the pixel level and frame level. Keywords Abnormal behavior detection · Real-time · Spatiotemporal autoencoder · Spatiotemporal convolutional neural network
1 Introduction Video surveillance data have increased dramatically in recent decades. With the ever-increasing emphasis on social security, the detection of abnormal crowd behavior has become an important and challenging task [1]. However, there are many kinds of abnormal behaviors in the real world and it is difficult to define them in all different scenarios. In additionally, most of the existing abnormal behavior detection algorithms have high computational complexity and cannot detect anomalies in real time, which limits their applications in the real world. The two problems above have become the bottleneck of the development in this field. The types of abnormal behavior are complex, and they are defined differently in distinct scenarios. For example, it
B
Zheyi Fan [email protected] Jianyuan Yin [email protected] Yu Song [email protected] Zhiwen Liu [email protected]
1
School of Information and Electronics, Beijing Institute of Technology, Beijing, China
is considered normal behavior for a vehicle to appear on a highway but an abnormal behavior if it appeared in a park. In view of this, the unsupervised learning is usually used in the existing detection model, which only learns normal behaviors and categorizes any behavior inconsistent to the norm as abnormal. Yong et al. [2] proposed the spatiotemporal autoencoder, the input of which is the video frame, to learn the normal behavior information to construct the abnormal behavior detection model, but the abnormal behavior cannot be located. In [3], t
Data Loading...
