Weakly Supervised Video Anomaly Detection Based on 3D Convolution and LSTM.

Area of Science:

• Computer Vision
• Machine Learning

Background:

• Weakly supervised video anomaly detection is a growing research area.
• Existing methods often focus on frame-level classification, limiting comprehensive anomaly detection.

Purpose of the Study:

• To propose a novel neural network architecture for efficient video anomaly detection.
• To treat videos as integral inputs for improved anomaly detection accuracy.

Main Methods:

• Utilizing three-dimensional convolutions for spatial and temporal feature extraction.
• Employing a Long Short-Term Memory (LSTM) network to model feature relationships.
• Implementing a video-label assignment procedure for detection.

Main Results:

• The proposed architecture demonstrates high computational efficiency.
• Extensive experiments confirm the effectiveness of the developed method.

Conclusions:

• The novel neural network architecture provides an efficient and effective solution for weakly supervised video anomaly detection.
• The approach advances the field by enabling integral video analysis rather than frame-level classification.