Autonomous vehicles (AVs) have emerged as captivating engineering ventures in the twenty-first century, capturing the interest of numerous academics and engineers across multiple generations. The world looks forward to leveraging AVs for reducing accidents caused by human errors and optimizing parking space utilization, particularly in urban areas. Accurate localization is pivotal for effective AV navigation, enabling the vehicle to pinpoint its precise position. While global positioning system (GPS) coordinates are widely used, their inherent errors and limitations can render them inadequate for determining precise location information, particularly in urban settings. Furthermore, drifting errors can undermine the efficacy of simultaneous localization and mapping (SLAM) algorithms. The proposed approach involves the utilization of a deep neural network, specifically a modified AlexNet architecture, which is a convolutional neural network (CNN), for localizing AVs in well-lit urban driving environments. The CNN enhances accuracy while reducing computational complexity and training time. Instead of relying on costly light detection and ranging (LiDAR) or radar sensors, a more affordable red green blue (RGB) camera sensor is employed. During testing, depth images are combined with RGB images using the intensity hue saturation (IHS) algorithm to enhance precision. Simulation results demonstrate an impressive accuracy rate of 95.49%, affirming the effectiveness of the proposed strategy. This study introduces a lightweight, precise, and reliable CNN architecture that significantly improves the accuracy of AV localization, simultaneously reducing predicted position errors by a considerable margin. The network's superiority is evidenced by mean square error (MSE) values of 0.039, 0.0099, and 0.0047 for position x, y, and orientation predictions, respectively. To validate real-time performance, the trained CNN was implemented in Python and integrated into the car learning to act (CARLA) simulator, enabling the online localization of a self-driving vehicle. This application successfully showcases the feasibility and efficacy of the proposed method.