LEAD-YOLO: A Lightweight and Accurate Network for Small Object Detection in Autonomous Driving

Area of Science:

• Computer Vision
• Artificial Intelligence
• Autonomous Systems

Background:

• Accurate small object detection is vital for autonomous driving safety.
• Existing methods often increase model complexity, hindering edge deployment.
• Balancing detection performance and computational efficiency is a key challenge.

Purpose of the Study:

• To propose LEAD-YOLO, a lightweight and efficient network for small object detection in autonomous driving.
• To enhance small object perception and context understanding without compromising efficiency.
• To achieve superior detection accuracy while maintaining computational efficiency for edge devices.

Main Methods:

• LEAD-YOLO integrates a Convolutional Gated Transformer (CGF) and Dilated Feature Fusion (DFF) in the backbone.
• A hierarchical feature fusion module (HFFM) is employed in the neck for guided feature aggregation.
• A shared feature detection head (SFDH) with shared modules and detail enhancement branches is used in the head.

Main Results:

• LEAD-YOLO achieved 3.8% and 5.4% mAP improvements on nuImages, with a 24.1% parameter reduction.
• On VisDrone2019, performance gains reached 7.9% and 6.4% for mAP@0.5 and mAP@[0.5:0.95].
• Demonstrated a significant balance between detection accuracy and model efficiency.

Conclusions:

• LEAD-YOLO effectively addresses the challenge of small object detection in autonomous driving.
• The proposed architecture offers a compelling solution for efficient edge deployment.
• LEAD-YOLO shows substantial potential for real-world autonomous driving applications.