Driverless cars have become a major research priority in the automotive industry. Due to the advancement in technology, the development of intelligent transportation systems are accelerated. Autonomous vehicles are able to independently perceive their surroundings and make real-time decisions; however, challenges such as adverse weather, occlusions, and latency still hinder safe deployment. Visual perception forms the foundation of any autonomous driving system (ADS), enabling robust scene understanding and decision-making. This study presents an extensive analysis of three core perception tasks - object detection, lane detection, and steering angle prediction - using state-of-the-art AI models. It evaluates advanced architectures including CNNs, LSTMs, and transformer-based networks, and benchmarks leading approaches such as YOLOv5, SSD, SCNN, and the NVIDIA steering framework across key datasets, including BDD100K, Udacity, and CARLA. The paper also examines industry case studies, the synergy between AI algorithms and sensor fusion (camera-radar-lidar), and identifies research gaps, emerging trends, and strategic pathways for developing reliable, scalable, and ethically aligned autonomous driving systems.
Sibin et al. (2026) studied this question.