With the advancement of autonomous vehicle technology, real-time object recognition and segmentation have become increasingly important for safety and security concerns. The semantic segmentation process assigns each pixel in a received image to a predefined class. Pixel-wise classification is another term for semantic segmentation. This paper presents an analysis of semantic segmentation algorithms, including SAM from Meta Research, FASTSAM from the Chinese Academy of Sciences, EfficientSAM from Facebook AI Research, which was trained on 2% of original datasets using dataset distillation and knowledge distillation techniques, and detectron2. A recent advancement called FastSAM and EfficientSAM claims to address the computation demands of real-time object detection while maintaining high accuracy. Detectron2, an extension of Detectron, is renowned for its ease of use, in addition to its flexibility and modularity. Studies in this paper have been conducted using CityScape [17] dataset along with real-time footage acquired by self-driving automobiles to evaluate the framework's capabilities. This experiment is conducted on a MacBook Air powered by the Apple M1 chip, which integrates several components, including the neural engine, graphics processing unit (GPU), CPU, and unified memory architecture, onto a single chip. Our evaluation encompasses key performance metrics, including detection accuracy, inference speed, computational efficiency, and memory footprint. The findings will assist several researchers working on object identification and segmentation in making informed decisions about selecting object detection technologies for self-driving cars. The findings contribute to advancing the current state of autonomous vehicle technology by identifying optimal choices based on specific requirements and constraints in object detection.