For autonomous vehicles, interpreting road scenes in the context of the driving environment is an important challenge. Real-time visual segmentation is crucial in self-driving applications. The processing overhead of semantic segmentation needs to be decreased in order to make it practical for embedded systems and autonomous vehicles. The modeluses image-level tag annotations to develop a dense pixel-level prediction model for semantic segmentation. These tags show whether certain classes are present in an image. Traffic signs, streets, people walking, trees, other cars, etc. are all depicted in the photographs. The initiative is based on autonomous or self-driving vehicles. In autonomous driving, self-driving cars must understand their surroundings. To extract regions with drivable roads, the suggested approach uses pixel-level segmentation. It then conducts a qualitative and quantitative study to show how well the proposed dataset and road detection work. The accuracy metric of the ResNet50 algorithm with different epoch sizesis compared. It aims to categorize each pixel in an image collected by a camera mounted on a moving vehicle into one of several possible images. Various stages of the procedure are applied to the collected photos to create a segmented image. It can distinguish between up to six classes. The above models offer real-time solutions as well as essential applications like road scene information, environmental awareness and understanding, and provide 85% accuracy.
Multi-class Pixel Level Segmentation for Drivable Road Detection
Lect. Notes Electrical Eng.
International Conference on Communications and Cyber Physical Engineering 2018 ; 2024 ; Hyderabad, India February 28, 2024 - February 29, 2024
Proceedings of the 6th International Conference on Communications and Cyber Physical Engineering ; Chapter : 81 ; 841-852
2024-02-05
12 pages
Article/Chapter (Book)
Electronic Resource
English