Anomaly detection based on local nearest neighbor distance descriptor in crowded scenes

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Anomaly detection in crowded scenes is challenging due to complex interactions.
• Existing methods often rely on low-level features, lacking contextual information.
• High-dimensional feature descriptors increase computational cost and storage requirements.

Purpose of the Study:

• To propose a novel Local Nearest Neighbor Distance (LNND) descriptor for anomaly detection.
• To leverage spatial and temporal contextual information for improved anomaly detection.
• To develop a computationally efficient and compact descriptor for crowded scenes.

Main Methods:

• Developed the Local Nearest Neighbor Distance (LNND) descriptor.
• Incorporated spatial and temporal contextual information into the descriptor.
• Evaluated the descriptor's effectiveness on benchmark datasets for anomaly detection.

Main Results:

• The LNND descriptor efficiently captures contextual information from multiple events.
• LNND offers a compact representation with lower dimensionality compared to traditional descriptors.
• LNND-based methods achieve comparable or better performance than state-of-the-art approaches with reduced processing.

Conclusions:

• The LNND descriptor is effective for anomaly detection in crowded scenes.
• LNND provides a computationally efficient and accurate alternative to existing methods.
• The descriptor's ability to incorporate context and reduce dimensionality offers significant advantages.