Anomaly Detection on the Edge Using Smart Cameras under Low-Light Conditions

Area of Science:

• Computer Vision
• Artificial Intelligence
• Smart City Technology

Background:

• Smart city surveillance systems increasingly use cameras in diverse environments.
• Low-light conditions significantly challenge anomaly detection and object recognition.
• Current methods struggle with accuracy and efficiency in poor visibility scenarios.

Purpose of the Study:

• To develop an image enhancement technique for low-light conditions.
• To improve object detection accuracy and reduce false positives in surveillance.
• To enable efficient anomaly detection in challenging visual environments.

Main Methods:

• Feature extraction from input images.
• A classifier selects optimal multi-enhancement networks and distinguishes light conditions.
• Object detection algorithm applied post-enhancement.
• Implementation on separate IoT-edge devices for distributed processing.

Main Results:

• Enhanced object detection accuracy in low-light environments.
• Mitigation of false positive detections.
• Nearly one-second response time across all processing stages on the ExDark database.
• Improved overall detection performance.

Conclusions:

• The proposed technique effectively enhances image quality in low-light conditions.
• Distributed IoT-edge processing ensures efficient and rapid anomaly detection.
• The research contributes to more reliable smart city surveillance and worker safety systems.