Scholarlink

<p>Accurate lane detection is essential for the navigation and safety of autonomous vehicles and Advanced Driver-Assistance Systems (ADAS). In this study, we present an effective Convolutional Neural Network (CNN)-based framework for detecting lanes and surrounding objects. The network is trained on a MATLAB-annotated dataset and incorporates architectural enhancements such as batch normalization, max pooling, and the integration of both convolutional and transposed convolutional layers. These design choices contribute to precise image reconstruction and robust detection performance. This study introduces a novel framework that combines lane detection with YOLOv3 for simultaneous real-time object recognition and vehicle speed estimation, enhancing situational awareness in autonomous driving. The system leverages a U-Net-inspired CNN architecture for accurate pixel-level lane segmentation across diverse and challenging road conditions. Additionally, temporal smoothing techniques are employed to maintain consistency and stability in frame-by-frame video analysis, ensuring reliable detection in dynamic environments. The methodology extends to video analysis by integrating the pre-trained You Only Look Once (YOLO) model with the MoviePy library to enable efficient frame-by-frame object detection. Additionally, Gradient-Weighted Class Activation Mapping (Grad-CAM) is used to enhance model interpretability by highlighting key regions influencing predictions. The model demonstrated strong performance with an accuracy of 0.99 and a macro-average F1-score of 0.99 on the custom dataset. While these metrics are encouraging, we acknowledge the need for benchmarking against standard public datasets to comprehensively assess performance relative to existing methods.</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

Real-time lane detection for autonomous driving & GRAD-CAM visualization

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Rajeev Kumar Gupta,
  • Devshree Jadeja,
  • Yatendra Sahu,
  • Nilesh Kunhare

摘要

Accurate lane detection is essential for the navigation and safety of autonomous vehicles and Advanced Driver-Assistance Systems (ADAS). In this study, we present an effective Convolutional Neural Network (CNN)-based framework for detecting lanes and surrounding objects. The network is trained on a MATLAB-annotated dataset and incorporates architectural enhancements such as batch normalization, max pooling, and the integration of both convolutional and transposed convolutional layers. These design choices contribute to precise image reconstruction and robust detection performance. This study introduces a novel framework that combines lane detection with YOLOv3 for simultaneous real-time object recognition and vehicle speed estimation, enhancing situational awareness in autonomous driving. The system leverages a U-Net-inspired CNN architecture for accurate pixel-level lane segmentation across diverse and challenging road conditions. Additionally, temporal smoothing techniques are employed to maintain consistency and stability in frame-by-frame video analysis, ensuring reliable detection in dynamic environments. The methodology extends to video analysis by integrating the pre-trained You Only Look Once (YOLO) model with the MoviePy library to enable efficient frame-by-frame object detection. Additionally, Gradient-Weighted Class Activation Mapping (Grad-CAM) is used to enhance model interpretability by highlighting key regions influencing predictions. The model demonstrated strong performance with an accuracy of 0.99 and a macro-average F1-score of 0.99 on the custom dataset. While these metrics are encouraging, we acknowledge the need for benchmarking against standard public datasets to comprehensively assess performance relative to existing methods.