Detecting Anomaly Event in Video Based on Generative Adversarial Network.

Area of Science:

• Computer Vision
• Machine Learning
• Deep Learning

Background:

• Video anomaly detection is a complex computer vision challenge.
• Current methods primarily optimize deep neural network architectures.
• A gap exists in exploring ensemble learning for this task.

Purpose of the Study:

• To propose a novel video anomaly detection method by combining ensemble learning and deep neural networks.
• To introduce an ensemble Generative Adversarial Network (GAN) approach for improved anomaly detection.

Main Methods:

• Developed an ensemble Generative Adversarial Network (GAN) with multiple generators and discriminators.
• Trained generators and discriminators collaboratively, enabling mutual feedback.
• Focused on modeling the distribution of normal video data.

Main Results:

• The ensemble GAN demonstrated superior performance in modeling normal data distributions compared to single GANs.
• Experimental results on public datasets confirmed significant performance improvements.
• The proposed method outperformed existing state-of-the-art video anomaly detection techniques.

Conclusions:

• Ensemble learning significantly enhances the performance of individual anomaly detection models.
• The proposed ensemble GAN is an effective approach for video anomaly detection.
• This method offers a promising direction beyond solely focusing on deep neural network structural design.