Temporal Improvement of Video Traffic Anomaly Detection: A Positioning Paper
Xinyue Zhang, Yuxuan Zhao, Ka Lok Man, Jeremy S. Smith, Young-Ae Jung, Yutao Yue · 2024
Internet of Things (IoT) technology can provide real-time public facilities data to Intelligent Transportation Systems (ITS), especially the surveillance cameras' video data. With the popularity of video surveillance systems, Video Anomaly Detection (VAD) has become more important in society and traffic management. However, the traditional, highly manual-dependent VAD method and the neural network model-based VAD method (such as Recurrent Neural Networks, RNN) face a significant challenge in temporal stream processing. This paper analyses current specific temporal challenges and proposes a Transformer-based spatial-temporal VAD model to alleviate the influence of temporal limitations. With the development of Transformer models, global information consideration and long-term relationship building have become more accessible, and the processing of temporal information in video data has become more efficient.