Scholarlink

Autonomous vehicles (AVs) are emerging due to their potential to reduce human error in transportation. Essential to AVs are subsystems for navigation and safety, including lane detection, often utilizing AI to analyze camera-captured images. AI performance hinges on training data quality and quantity, which may be lacking in atypical driving scenarios. To address this, we explore the use of generative adversarial networks (GANs) for generating realistic training data. GANs employ two networks, a generator and a discriminator, in a mutually challenging training process. Our study uses driver-view images to train a MATLAB-based GAN, which successfully generates diverse, realistic images, as demonstrated by our tests.

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

Generative Adversarial Network Driver View Image Synthesis for Autonomous Vehicle Training

  • 章节

  • 发表日期：

    2026年

查看全文
  • Athanasius Joseph,
  • Adizul Ahmad,
  • Ihsan Mohd Yassin,
  • Azlee Zabidi,
  • Rajeswari Raju,
  • Megat Syahirul Amin Megat Ali

摘要

Autonomous vehicles (AVs) are emerging due to their potential to reduce human error in transportation. Essential to AVs are subsystems for navigation and safety, including lane detection, often utilizing AI to analyze camera-captured images. AI performance hinges on training data quality and quantity, which may be lacking in atypical driving scenarios. To address this, we explore the use of generative adversarial networks (GANs) for generating realistic training data. GANs employ two networks, a generator and a discriminator, in a mutually challenging training process. Our study uses driver-view images to train a MATLAB-based GAN, which successfully generates diverse, realistic images, as demonstrated by our tests.