基于全景图像的车位检测方法研究

文章摘要

基于全景图像的车位检测方法研究

Research on Parking Space Detection Method Based on Panoramic Images

投稿时间：2023-09-13 修订日期：2023-09-13

DOI：

中文关键词: YOLOv5s 车位检测 Ghost模块 SimSPPF模块

英文关键词: YOLOv5s Parking space detection Ghost module SimSPPF module

基金项目:安徽省科技重大专项“智能储电式胶轮电车研发与应用”（202103a05020033）

作者	单位	邮编
周晋伟^*	安徽工程大学	241000

摘要点击次数: 1043

全文下载次数: 0

中文摘要:

针对自动泊车过程中车位检测网络复杂，边缘部署困难等问题，提出了一种基于全景图像的车位检测方法。使用Ghost模块对其参数进行量化处理，同时引入EIoU损失函数，对检测尺度进行裁剪，降低计算成本，添加SimSPPF模块来取代YOLOv5s主干中的SPPF，以提高计算效率和准确的目标检测能力。实验结果表明，本文车位检测网络mAP@0.5达到98.2%，参数量、浮点运算量、权重分别降低了86.7%、62.5%、85.4%，检测速度提高了20%，车位检测准确率达到98.1%。

英文摘要:

A parking space detection method based on panoramic images is proposed to address the challenges of complex networks and difficulties in edge deployment during the automated parking process. The method incorporates the Ghost module for quantization of parameters and introduces the EIoU loss function to crop the detection scale, reducing computational costs. Additionally, the SimSPPF module is integrated to replace the SPPF module in the YOLOv5s backbone, enhancing computational efficiency and accurate object detection capabilities. Experimental results demonstrate that the proposed parking space detection network achieves an mAP@0.5 of 98.2%. The parameter count, floating-point operations, and weights are reduced by 86.7%, 62.5%, and 85.4%, respectively. The detection speed is increased by 20%, and the parking space detection accuracy reaches 98%.

View Fulltext 查看/发表评论下载PDF阅读器

关闭