Scholarlink

This paper presents a comprehensive software model for an autonomous vehicle system capable of detecting and navigating objects in real-time without human intervention. Our approach integrates CNN, computer vision, and camera-based perception into a sensor fusion system, inspired by Udacity’s self-driving car project and Nvidia’s “PilotNet.” The model emphasizes real-time responsiveness and computational efficiency, especially on low-power hardware like the Raspberry Pi. It processes camera data to make driving decisions, calculating steering angles, adjusting speed, and selecting optimal trajectories. Using a deep learning approach, convolutional neural networks (CNNs) are trained to map raw camera data directly to control commands, ensuring smooth and accurate steering angle predictions. A smoothness constraint in the loss function minimizes jerkiness, improving real-world driving behavior. Simulations validate the model’s performance, demonstrating the feasibility of autonomous driving technology. The paper also discusses future advancements, trends, challenges, and open research questions in the field of autonomous vehicles.

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

Autonomous Vehicle Steering Angle Prediction Using CNN and Computer Vision

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Abhinav Bhatt,
  • Amith Vikram Rajaram,
  • K. B. Surekha,
  • Geeta Amol Patil,
  • B. V. Rithika,
  • Aditya V. Patil

摘要

This paper presents a comprehensive software model for an autonomous vehicle system capable of detecting and navigating objects in real-time without human intervention. Our approach integrates CNN, computer vision, and camera-based perception into a sensor fusion system, inspired by Udacity’s self-driving car project and Nvidia’s “PilotNet.” The model emphasizes real-time responsiveness and computational efficiency, especially on low-power hardware like the Raspberry Pi. It processes camera data to make driving decisions, calculating steering angles, adjusting speed, and selecting optimal trajectories. Using a deep learning approach, convolutional neural networks (CNNs) are trained to map raw camera data directly to control commands, ensuring smooth and accurate steering angle predictions. A smoothness constraint in the loss function minimizes jerkiness, improving real-world driving behavior. Simulations validate the model’s performance, demonstrating the feasibility of autonomous driving technology. The paper also discusses future advancements, trends, challenges, and open research questions in the field of autonomous vehicles.