Joint Generative Network for Abnormal Event Detection in Surveillance Videos
Savath Saypadith, Sunepha Detvongsa, Takao Onoye · 2022 19th International SoC Design Conference (ISOCC) · 2022
Abnormal events can be seen as spatiotemporal objects. Recent methods attempt to extract features from spatial information and learn abnormal events with convolutional neural networks. However, such representations lack motion information to model abnormal events. In this paper, we propose an architecture based on generative network structure to learn and detect abnormal events. Spatial and temporal encoders are employed to extract features from images. Shortcut Inception Modules (SIM) are used in encoders to keep meaningful information during training the neural network, which also reduces the network parameters. Experimental results on benchmark datasets show that the architecture performed comparative results in terms of detection accuracy and processing time compared to learning-based methods.