A Vision-Based Lane Detection Approach for Autonomous Vehicles Using a Convolutional Neural Network Architecture

Autonomous vehicles no longer belong to the realm of science fiction. They have become a prominent area of research in the last two decades because of the integration of Artificial Intelligence in the automobile industry. Apart from the development of various complex learning algorithms, the advancement of cameras, sensors, and geolocation technology as well as the escalation in the capacity of machines have played a crucial role in bringing this technology into reality. We have had significant breakthroughs in the development of autonomous cars within the last ten years. However, despite the success of multiple prototypes in navigating within the borders of a delimited area, researchers are yet to overcome several drawbacks before embodying them in the transport system; and one of those hurdles lies in the lane detection system of the cars. Therefore, in this article, we present an intelligent lane detection algorithm incorporating fully-connected Neural Networks with a secondary layer protection scheme to detect the borders of a lane. We achieved over 98% classification accuracy using the proposed lane detection model. We also implemented the model in a small prototype to take a look at its performance. Experimental results infer that the algorithm is capable of lane detection and ready for practical use.