Object detection model of vehicle-road cooperative autonomous driving based on improved YOLO11 algorithm

Area of Science:

• Computer Vision
• Artificial Intelligence
• Autonomous Systems

Background:

• Autonomous driving systems face challenges in accurately detecting small, occluded targets in complex roadside and distant scenes.
• Existing algorithms struggle with low detection accuracy, false positives, and missed detections.

Purpose of the Study:

• To develop an improved small target detection algorithm for autonomous driving based on YOLOv11.
• To enhance feature extraction, small target understanding, and model efficiency in complex scenarios.

Main Methods:

• Embedded Channel Transposed Attention (C3CTA module) in the Backbone network to improve feature extraction and occlusion handling.
• Introduced Diffusion Focusing Pyramid Network in the Neck to enhance small target understanding in complex scenes.
• Implemented a Lightweight Shared Convolutional Detection Head to reduce model parameters for lightweight requirements.

Main Results:

• The improved algorithm achieved Precision of 85.7%, Recall of 79.4%, mAP@0.5 of 85.3%, and mAP@0.5-95 of 61.3% on the DAIR-V2X-I dataset.
• Demonstrated significant improvements over the baseline model, with increases of 4.0%, 3.1%, 2.4%, and 2.8% in key metrics.
• Successfully addressed challenges of overlapping occlusion and improved the extraction of vehicle target features.

Conclusions:

• The proposed YOLOv11-based algorithm effectively enhances small target detection accuracy in autonomous driving.
• The integration of C3CTA, Diffusion Focusing Pyramid Network, and Lightweight Shared Convolutional Detection Head proves effective.
• This work offers a novel solution for robust target detection in challenging autonomous driving environments.