FRCP-YOLO: Road object detection algorithm based on improved YOLOv8n.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Autonomous Systems

Background:

• Accurate road object detection is vital for autonomous vehicle safety.
• Current models face challenges with small objects, low accuracy, and poor robustness.

Purpose of the Study:

• To propose FRCP-YOLO, an enhanced road object detection model based on YOLOv8n.
• To improve detection accuracy, reduce model complexity, and enhance robustness, especially for small objects.

Main Methods:

• Replaced C2f module with FasterNet Block for faster feature extraction.
• Introduced R-CA module for improved object focus and feature learning.
• Implemented a high-resolution branch and detection head for small object detection.
• Utilized PIoU v2 loss function for precise bounding box regression.

Main Results:

• FRCP-YOLO achieved 0.924 mAP@50 and 0.667 mAP@50-95 on the KITTI dataset, outperforming the baseline by 5.0% and 6.6%.
• Reduced model parameters by 4% compared to the baseline.
• Demonstrated superior performance on the BDD100K dataset in complex scenarios like dense traffic and low light.

Conclusions:

• FRCP-YOLO offers improved accuracy, efficiency, and robustness for road object detection.
• The model shows strong generalization capabilities, making it reliable for autonomous driving in diverse conditions.