Real-time adversarial GAN-based abnormal crowd behavior detection
- PDF / 925,235 Bytes
- 10 Pages / 595.276 x 790.866 pts Page_size
- 26 Downloads / 210 Views
SPECIAL ISSUE PAPER
Real‑time adversarial GAN‑based abnormal crowd behavior detection Qiulei Han1 · Haofeng Wang2 · Lin Yang2 · Min Wu2 · Jinqiao Kou2 · Qinsheng Du1 · Nianfeng Li1 Received: 25 March 2020 / Accepted: 6 October 2020 © Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract Detecting abnormal events in the crowd is a challenging problem. Insufficient samples make those traditional model-based methods cannot cope with sophisticated anomaly monitoring. Therefore, we design a real-time generative adversarial network plus an add-on encoder to deal with the continually changing environment. After the generator reconstructs the compressed pattern to generate the design to the latent vector, a discriminator is used to construct better videos by minimizing the adversarial loss function. We calculated the abnormal score by the distance between the two underlying patterns encoded by the first and the second encoders. The unusual event is detected when the anomaly score is above the threshold. To accelerate the processing efficiency, we introduced the grouped pointwise convolution method to decrease the computing complexity. The frame-level and video-level experiments on the benchmark dataset show the accuracy and reliance of our approach. The acceleration approach can increase the efficiency of the network with only limited accuracy loss. Keywords Generative adversarial network (GAN) · Anomaly detection · Auto-encoder · Grouped pointwise convolution
1 Introduction Public places use more and more surveillance cameras, e.g., public transportation systems, hospitals, shopping malls, parks, etc. The enormous safety cameras create a massive amount of videos, and potential application includes object detection, tracking, image retrieval, and so on. At the same time, limited human monitoring cannot keep up with the demand for surveillance. Automated alarming abnormal events is an urgent problem to conquer. In surveillance videos, suspicious events are challenging to describe and predict because of the intravariance and inter-similarity. We assume that regular competition indicates commonality. Therefore, an unusual event is associated with the distinctness of the activity. For example, we interpret a running person in a everybody-walks scene as abnormal. Unfortunately, due to the elaborate scenes and the uncertainty of anomaly, this abnormal detection is still challenging and hard to be distinguished. * Qiulei Han [email protected] 1
School of Computer Science, Changchun University, Changchun 130022, People’s Republic of China
Beijing Institute of Computer Technology and Applications, Beijing 110000, People’s Republic of China
2
In relation to abnormal activity detection, numerous studies declare they can solve the problem in different application environments. For unsupervised and semi-supervised problems, a generative adversarial network (GAN) has become a representative method in the artificial intelligence field. In the traditional network, the high dimensional vector needs to be trans
Data Loading...
