YOLO-IR-Free: An Improved Algorithm for Real-Time Detection of Vehicles in Infrared Images

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Infrared vehicle detection is crucial for traffic safety and intelligent driving systems, especially in adverse conditions.
• Existing methods struggle with low contrast, small object detection, and real-time performance in infrared imaging.
• Lightweight object detection algorithms often face a trade-off between speed and accuracy for this task.

Purpose of the Study:

• To develop an improved, real-time infrared vehicle detection algorithm.
• To address limitations of current methods in handling low contrast and small objects.
• To enhance the balance between detection speed and accuracy in lightweight object detection.

Main Methods:

• Proposed YOLO-IR-Free, an anchor-free algorithm based on an improved attention mechanism YOLOv7.
• Introduced a novel attention mechanism and network module to capture subtle textures and low-contrast features.
• Replaced anchor-based detection head with an anchor-free head to improve detection speed.

Main Results:

• YOLO-IR-Free demonstrated superior performance compared to other methods.
• Achieved higher accuracy, recall rate, and average precision scores.
• Maintained effective real-time detection performance.

Conclusions:

• YOLO-IR-Free effectively tackles challenges in real-time infrared vehicle detection.
• The proposed attention mechanism and anchor-free design significantly improve performance.
• This algorithm offers a promising solution for enhanced traffic safety and intelligent driving.