A Comprehensive Review of Image Segmentation Architectures for Autonomous Navigation and Intelligent Transportation Systems

Segmentation architectures are essential in computer vision, enabling the precise extraction of meaningful information from images and videos by breaking them down into understandable parts. This paper explores six key segmentation architectures: Fully Convolutional Network (FCN), U-Net, DeepLab, YOLO (You Only Look Once) and Mask R-CNN, focusing on their unique methods, strengths, and applications, especially in autonomous navigation. In autonomous vehicles, tasks like lane detection, road segmentation, and identifying objects such as pedestrians and other vehicles rely heavily on semantic segmentation, instance segmentation, and object detection. FCN and U-Net are great at semantic segmentation, making them suitable for understanding the overall scene. DeepLab stands out in tasks that need high-resolution and fine detail. YOLO is preferred when for dynamic environments. This review aims to help researchers and practitioners choose the right architecture based on their specific needs, such as available computational resources, required precision, and the need for real-time processing. By doing so, it seeks to advance the development of smarter transportation systems and better autonomous navigation technologies.