LPCF-YOLO: A YOLO-Based Lightweight Algorithm for Pedestrian Anomaly Detection with Parallel Cross-Fusion

Area of Science:

• Computer Vision
• Deep Learning
• Artificial Intelligence

Background:

• Current pedestrian anomaly detection networks are often too complex for practical deployment.
• High computational requirements hinder the real-world application of advanced detection models.

Purpose of the Study:

• To develop a lightweight anomaly detection network for efficient pedestrian monitoring.
• To reduce the parameter count and computational load of existing models without sacrificing performance.

Main Methods:

• Proposed LPCF-YOLO (Lightweight Parallel Cross-Fusion YOLO) based on YOLOv8n.
• Introduced FPC-F and S-EMCP modules in the backbone, and an ADown module for cost reduction.
• Implemented a Lightweight High-level Screening Feature Pyramid Network (L-HSFPN) in the neck.
• Utilized Wise-IoU loss for improved localization and generalization.

Main Results:

• Reduced parameters by 30.33% and FLOPs by 79.01% compared to YOLOv8n.
• Achieved 2.09 M parameters and 1.7 G FLOPs, with a 179.62% increase in FPS to 43.9.
• Maintained or slightly improved mean average precision (mAP@0.5) on UCSD-Ped1 and UCSD-Ped2 datasets.

Conclusions:

• LPCF-YOLO offers a significantly more efficient alternative for pedestrian anomaly detection.
• The proposed lightweight architecture enables practical real-world deployment of anomaly detection systems.