Semantic Segmentation Using U-Net for Autonomous Driving
Venkat Sai Natte, Jahnavi Kadicherla, Sri Divya Katukojwala, K. Swaraja, N. Arun Vignesh, Srilakshmi Aouthu · 2024
Safe autonomous driving crucially depends on effective detection of nearby objects for hazard-free navigation. Semantic segmentation aids autonomous systems by enabling accurate environmental perception. However, identifying moving vehicles in images, especially when obscured, poses challenges. To address this, deep convolutional neural networks (CNNs), including UNET, are employed. By integrating CNNs' information across different scales into images of varying resolutions, the method enhances vehicle detection accuracy. Thorough training on real-world data refines the model, demonstrating quick and precise vehicle identification. Experimental results highlight significant improvements in detection accuracy (81.4%) and mean intersection over union (76.84%). This approach not only advances real-time vehicle detection but also emphasizes UNET's adaptability in dynamic traffic environments. The findings suggest a robust pathway for deploying dependable detection systems, enhancing overall autonomous driving safety.