CNN-Based Person Detection Using Infrared Images for Night-Time Intrusion Warning Systems

Area of Science:

• Computer Vision
• Artificial Intelligence
• Surveillance Technology

Background:

• Night-time surveillance is crucial for public safety and security.
• Existing methods for detecting people in infrared CCTV (closed-circuit television) images face challenges due to overhead camera placement and small human target regions.
• Automated detection of intruders in restricted areas using infrared cameras is an active research area.

Purpose of the Study:

• To propose an accurate and efficient method for detecting people in infrared CCTV images during night-time.
• To address the challenges of detecting small human figures from overhead infrared cameras.
• To improve the performance of human detection in infrared imagery for enhanced security.

Main Methods:

• Construction of three infrared image datasets from public beach CCTV and a pedestrian bridge FLIR (forward-looking infrared) camera.
• Implementation of a convolution neural network (CNN)-based pixel-wise classifier for fine-grained person detection.
• Comparative analysis of the proposed CNN method against five conventional detection techniques.

Main Results:

• The proposed CNN-based human detection method demonstrated superior performance compared to conventional approaches across all tested datasets.
• The method achieved F1 scores exceeding 80% for object-level detection in all datasets.
• Significant improvement in the accuracy and efficiency of person detection from infrared images was observed.

Conclusions:

• The developed CNN-based approach offers a more effective solution for night-time person detection in infrared imagery.
• This research contributes to enhancing the safety and security of public areas through improved automated surveillance.
• The findings highlight the potential of deep learning for robust detection in challenging surveillance scenarios.