Abstract: Motion objects in video frames often exhibit diverse scales over time, which poses a challenge for video anomaly detection. Although traditional generative adversarial networks (GANs) have achieved some success in video anomaly detection tasks, their performance is limited due to the use of a single-scale feature extraction that fails to capture features of objects at different scales. To address this issue, this paper proposes a video anomaly detection method based on a GAN structure that incorporates multi-scale feature fusion and attention mechanisms. Specifically, different-sized convolutional kernels are employed to capture features with varying receptive fields, which are then fused to obtain multi-scale feature representations. Additionally, a coordinate attention mechanism is introduced after the transposed convolutional layers of the generator, allowing adaptive allocation of feature map weights to enhance the model’s perception of crucial features.Experimental results on the public datasets UCSD Ped2 and Avenue demonstrate that the proposed method outperforms existing approaches.

Key words: video anomaly detection, deep learning, generative adversarial networks, multi-scale feature fusion, attention mechanism