Deep anomaly detection in expressway based on edge computing and deep learning
- PDF / 4,001,904 Bytes
- 13 Pages / 595.276 x 790.866 pts Page_size
- 85 Downloads / 233 Views
ORIGINAL RESEARCH
Deep anomaly detection in expressway based on edge computing and deep learning Juan Wang1 · Meng Wang1 · Qingling Liu2 · Guanxiang Yin1 · Yuejin Zhang1 Received: 2 June 2020 / Accepted: 22 September 2020 © Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract In order to improve the real-time efficiency of expressway operation monitoring and management, the anomaly detection in intelligent monitoring network of expressway based on edge computing and deep learning is studied. The video data collected by the camera equipment in the intelligent monitoring network structure of the expressway is transmitted to the edge processing server for screening and then sent to the convolutional neural network. The convolutional neural network uses the multi-scale optical flow histogram method to preprocess the video data after the edge calculation to generate the training sample set and send it to the AlexNet model for feature extraction. SVM classifier model is used to train the feature data set and input the features of the test samples into the trained SVM classifier model to realize the anomaly detection in the intelligent monitoring network of expressway. The research method is used to detect the anomaly in an intelligent monitoring network of an expressway. The experimental results show that the method has better detection effect. The miss rate has reduced by 20.34% and 40.76% on average compared with machine learning method and small block learning method, respectively. The false positive rate has reduced by 27.67% and 21.77%, and the detection time is greatly shortened. Keywords Edge computing · Deep learning · Intelligent monitoring · Anomaly detection · AlexNet network
1 Introduction Building a highly intelligent highway video surveillance network and network video anomaly detection are important components of future intelligent transportation systems (Li et al. 2016), as shown in Fig. 1. They aid traffic monitoring, counting, and surveillance, which are necessary for tracking the performance of traffic operations (Zhou et al. 2016). Now cities are basically installing field devices such as fixed-position cameras or motion sensors on traffic lights to monitor vehicles (Mansour et al. 2019). However, the image or video of the vehicle target will inevitably change due to factors such as light, viewing angle and vehicle interior. With the rapid development of deep learning theory Juan Wang and Meng Wang contributed equally. * Yuejin Zhang [email protected] 1
School of Information Engineering, East China Jiaotong University, Nanchang 330013, China
College of Information and Communication Engineering, Harbin Engineering University, Harbin, China
2
and practice in recent years, researchers at home and abroad have found that deep convolutional neural networks have a certain degree of invariance to geometric transformation, deformation and lighting, and can effectively overcome the variability of vehicle appearance, and the feature description is adapted to the training data to
Data Loading...
