YOLOv8-QSD: An Improved Small Object Detection Algorithm for Autonomous Vehicles Based on YOLOv8

As self-driving vehicles become more prevalent, the speed and accuracy of detecting surrounding objects through onboard sensing technology have become increasingly important. The YOLOv8-QSD network is a novel anchor-free driving scene detection network that builds on YOLOv8 and ensures detection accuracy while maintaining efficiency. The network’s backbone employs structural reparameterization techniques to transform the diverse branch block (DBB)-based model. To accurately detect small objects, it integrates features of different scales and implements a bidirectional feature pyramid network (BiFPN)-based feature pyramid after the backbone. To address the challenge of long-range detection in driving scenarios, a query-based model with a new pipeline structure is introduced. The test results demonstrate that this algorithm outperforms YOLOv8 on the large-scale small object detection dataset (SODA-A) in terms of both speed and accuracy. With an accuracy rate of 64.5% and reduced computational requirements of 7.1 GFLOPs, it satisfies the speed, precision, and cost-effectiveness requirements for commercial vehicles in high-speed road driving scenarios.

I. Introduction

The autonomous driving landscape has undergone a remarkable transformation due to the significant advancement in autonomous driving perception technology, the continuous progress in deep learning and enhanced computational capabilities, and the availability of cost-effective onboard sensors, such as cameras, radars, and LiDARs. To ensure the seamless operation of autonomous commercial vehicles in complex traffic scenarios, it is necessary to develop a robust perception system with the following essential characteristics [1], [2], [3]. 1)

Accuracy: The perception system must accurately identify the categories and sizes of nearby objects and predict their behavior in complex traffic scenarios. For high-speed commercial vehicles on highways, it is crucial to improve the perception of distant small objects. The detection capability of small objects is severely tested.

2)

Real-Time Performance: The perception task should achieve high accuracy while completing it as quick as possible. Real-time performance is critical because high-latency perception systems can cause delays in vehicle decision-making and control [4], potentially resulting in fatal accidents. Lightweight models appropriate for onboard perception modules are crucial, necessitating low computational power and cost, as well as strong processing capabilities for small objects.

3)

Robustness: The perception system must maintain normal operational performance in adverse environmental conditions, such as occlusions and low-light situations, ensuring the safety and stability of autonomous driving [5].