Scholarlink

<p>To meet the increased parking challenges posed by the rapid growth in university parking demand, there is a growing demand for smart parking systems to alleviate stress on access, congestion and optimise space. In this paper we present an Automated Vehicle Access Control (AVAC) created by deep learning (DL) based License Plate Recognition (LPR) of vehicles. The AVAC used for vehicle access uses the InceptionResNetV2 architecture for plate detection with an Intersection over Union (IoU) performance metric of 0.815 and a F1 score of 0.973 as well as a hybrid of the MobileNetV2 – EfficientNetV2 architecture designed for character recognition with 98.59% accuracy and 1.41% Character Error Rate (CER). The AVAC was trained using an Indian vehicle dataset containing over 3000 images per character class and processes images at a resolution of 224 <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\times\)</EquationSource> </InlineEquation> 224 with a mean inference time of &lt;50 ms on NVIDIA T4 GPUs. The AVAC has demonstrated its capability of providing real time access control; thus, leading to streamlined vehicle authentication, assigning parking places and access control while simultaneously minimising traffic congestion and operational cost.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automated vehicle access control system using license plate recognition

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月12日

查看全文
  • R. Advaith,
  • Venkat K. Uditya,
  • S. Hariharan,
  • R. G. Sangeetha

摘要

To meet the increased parking challenges posed by the rapid growth in university parking demand, there is a growing demand for smart parking systems to alleviate stress on access, congestion and optimise space. In this paper we present an Automated Vehicle Access Control (AVAC) created by deep learning (DL) based License Plate Recognition (LPR) of vehicles. The AVAC used for vehicle access uses the InceptionResNetV2 architecture for plate detection with an Intersection over Union (IoU) performance metric of 0.815 and a F1 score of 0.973 as well as a hybrid of the MobileNetV2 – EfficientNetV2 architecture designed for character recognition with 98.59% accuracy and 1.41% Character Error Rate (CER). The AVAC was trained using an Indian vehicle dataset containing over 3000 images per character class and processes images at a resolution of 224 \(\times\) 224 with a mean inference time of <50 ms on NVIDIA T4 GPUs. The AVAC has demonstrated its capability of providing real time access control; thus, leading to streamlined vehicle authentication, assigning parking places and access control while simultaneously minimising traffic congestion and operational cost.