Driverless Image Processing Based on Improved YOLOv5l | IEEE Conference Publication | IEEE Xplore

Driverless Image Processing Based on Improved YOLOv5l


Abstract:

In the field of autonomous driving, road information is typically assessed and processed using a combination of 3D and 2D target detection algorithms. Among these, the YO...Show More

Abstract:

In the field of autonomous driving, road information is typically assessed and processed using a combination of 3D and 2D target detection algorithms. Among these, the YOLO series algorithms stand out in the domain of 2D target detection. This paper aims to enhance YOLOv5l and apply it to the field of driverless image processing to improve operational efficiency. Addressing the issues of parameter quantity and operational efficiency in the traditional YOLOv5l algorithm, we propose an improvement strategy. Firstly, we incorporate the lightweight network ShuffleNetV2 to enhance network feature extraction efficiency. Secondly, we introduce the lightweight convolution structure GSConv to reduce the computational load of the model. Lastly, we introduce SIoU loss to optimize the regression model. Experimental results demonstrate that our improved algorithm achieves comparable detection accuracy to the traditional YOLOv5l algorithm on the KITTI dataset, while also exhibiting higher detection efficiency and fewer parameters. These findings validate the effectiveness and practicality of our improved algorithm. This enhancement offers valuable insights for further research on object detection algorithms in the field of autonomous driving.
Date of Conference: 10-12 November 2023
Date Added to IEEE Xplore: 16 May 2024
ISBN Information:
Conference Location: Jiaxing, China

Funding Agency:


I. Introduction

According to a technical report by the National Highway Traffic Safety Administration (NHTSA), 94% of road accidents are caused by human faults [1]. In recent years, the rapid development of computer vision and deep learning technology has led to significant breakthroughs in computer vision, making driverless technology an increasingly popular research direction in the field of autonomous driving. In this context, the development of driverless systems holds great promise in preventing accidents, reducing emissions, providing transportation for individuals with limited mobility, and alleviating the stress associated with driving. Driverless technology has vast potential applications that can bring about substantial changes in transportation, logistics distribution, and urban planning. However, the precise real-time detection and tracking of objects are of utmost importance for ensuring efficient and accurate driverless systems.

References

References is not available for this document.