GridHTM: Grid-Based Hierarchical Temporal Memory for Anomaly Detection in Videos.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Video anomaly detection is crucial for surveillance but faces challenges with deep learning methods like noise, concept drift, and data requirements.
• Existing deep learning approaches, mainly generative models, struggle with inherent anomaly detection complexities such as unknown anomalies and class imbalance.

Purpose of the Study:

• To explore the Hierarchical Temporal Memory (HTM) algorithm for video anomaly detection.
• To introduce and evaluate GridHTM, a novel grid-based HTM architecture designed for complex video surveillance analysis.

Main Methods:

• Exploration of Hierarchical Temporal Memory (HTM) algorithm's suitability for video anomaly detection.
• Development and testing of GridHTM, a specialized grid-based HTM architecture.
• Evaluation using the VIRAT video surveillance dataset.

Main Results:

• GridHTM demonstrates favorable properties like noise tolerance and online learning, addressing concept drift.
• The system shows great potential for real-time unsupervised anomaly detection in complex video data.
• Evaluation results validate the effectiveness of GridHTM on the VIRAT dataset.

Conclusions:

• Hierarchical Temporal Memory (HTM) offers a promising alternative to deep learning for video anomaly detection.
• GridHTM is a viable solution for real-time, unsupervised anomaly detection in challenging surveillance environments.
• The system's online learning capability is key to adapting to evolving video data patterns.