Scholarlink

<p>Autonomous cars and intelligent transportation systems depend on the precise recognition of time-dependent traffic signals in real time. The accuracy of traditional traffic sign identification techniques is frequently reduced by differences in illumination, weather, and sign deterioration over time. The proposed framework in the given study overcomes these limitations through a multi-stage process that incorporates a traffic sign recognition dataset and adapts its feature extraction and classification layers accordingly. The proposed framework involves augmentation of the image dataset followed by feature extraction using attention-based convolutional neural network (CNN) architecture. It is subjected to feature selection and tuning using the gravity kinematic optimization (GKO) approach. GKO dynamically adjusts the attention weights, thereby filtering out irrelevant information and improving model convergence. For visualization of the output results, YOLOv4 (object detection model) is used. It uses residual blocks to achieve real-time classification results without increasing detection time. Experimental results show that the proposed framework achieves the highest training accuracy (99.09%), validation accuracy (98.12%), and the lowest frames per second (FPS 23.91), thus enhancing the safety and reliability of transportation systems.</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

CNN-GKO-YOLOv4: an optimized neural network framework for dynamic traffic sign recognition

  • 期刊论文

  • 发表日期：

    2025年12月27日

查看全文
  • Vinay P,
  • Khan Vajid Nabilal,
  • Harshitha K M,
  • Geetanjali Sharma,
  • Rajashree T. Gadhave,
  • Sreedhar Bhukya

摘要

Autonomous cars and intelligent transportation systems depend on the precise recognition of time-dependent traffic signals in real time. The accuracy of traditional traffic sign identification techniques is frequently reduced by differences in illumination, weather, and sign deterioration over time. The proposed framework in the given study overcomes these limitations through a multi-stage process that incorporates a traffic sign recognition dataset and adapts its feature extraction and classification layers accordingly. The proposed framework involves augmentation of the image dataset followed by feature extraction using attention-based convolutional neural network (CNN) architecture. It is subjected to feature selection and tuning using the gravity kinematic optimization (GKO) approach. GKO dynamically adjusts the attention weights, thereby filtering out irrelevant information and improving model convergence. For visualization of the output results, YOLOv4 (object detection model) is used. It uses residual blocks to achieve real-time classification results without increasing detection time. Experimental results show that the proposed framework achieves the highest training accuracy (99.09%), validation accuracy (98.12%), and the lowest frames per second (FPS 23.91), thus enhancing the safety and reliability of transportation systems.