The rise of vehicle autonomous driving technology has surged lane detection into a key area of research and development. Conventional lane detection methodologies, reliant on feature extraction and high-definition imagery, grapple with computational demands. Detecting lanes accurately requires considering their continuous nature and leveraging information from previous frames to improve accuracy. To address these challenges, we propose a hybrid deep learning-based approach using a combination of fully connected CNNs, pre-trained models like ResNet, and LaneNet architecture. This model is designed to identify and locate lanes within images, with extensive experiments conducted on large-scale datasets demonstrating the effectiveness of our method. A groundbreaking lane fitting process, incorporating a dynamic variable matrix, elevates detection efficacy, particularly in scenarios characterized by variable slopes. The resultant real-time performance amplification is nothing short of remarkable. Comprehensive experimentation conducted across an extensive array of scenarios bears witness to the unparalleled efficiency of our method, transcending the capabilities of existing models. Our approach shines brightly in the darkest of conditions: heavy shadows, deteriorated lane markings, and even vehicular obfuscation. This paradigm shift in lane detection promises a revolution in the domain of autonomous vehicles and driver assistance systems, redefining safety and reliability standards across the industry.