This paper is aimed at accelerating the development of fully autonomous vehicles by identifying objects around them with great precision. Semantic segmentation, as described by Li et al. [1], is crucial in autonomous vehicles as it enables real-time understanding of the environment by classifying every pixel in an image, allowing the vehicle to accurately identify objects like roads, pedestrians, and obstacles for navigation enhancing road safety. By comparing various semantic segmentation models trained mainly on 512x512 resolution images, we narrowed down to Segformer-B0, a vision transformer model, due to its better performance. In order to attain real-time processing, the model was optimized, which led to an increase of 74 percent in frames per second (fps). However, porting the optimized model to a constrained embedded platform like Jetson Nano still resulted in low fps. Moreover, we present an idea that applies IoT control systems, which facilitates entering human control in the case of failure during self-driving. IoT control system is tailored for operation with a Wi-Fi enabled microcontroller to guarantee full connectivity and control. This holistic framework incorporates advanced technologies like AI, IoT, and embedded systems, thereby furthering the development of autonomous vehicles with advanced perception and safety features.