Abstract
Lane detection is a critical component of autonomous vehicles, and essential for ensuring their safe and efficient navigation. In recent years, several models have been proposed to improve the accuracy and performance of this activity, however, little is known about practical real-time implementations of these models in small-scale vehicles and low-cost devices, with limited computing and memory resources. This work presents an overview of the models and techniques used for lane recognition in autonomous vehicles. Additionally, an experimental evaluation was conducted for testing different models for lane detection in real-time on miniature cars, identifying the best-performing models. Three methods were analysed: (1) a model using traditional image processing techniques, (2) an end-to-end deep learning (DL) model proposed by Nvidia, which directly predicts angles, and (3) an ultra-fast lane detection DL model, which predicts lane boundaries. In summary, the results demonstrate the adequacy of DL models for addressing the lane detection recognition, even on low-cost devices, as long as minimum capacity and memory requirements are met. It was observed that the ultra-fast lane detection DL model outperformed the other 2 methods with predicting accuracy of 95% compared to 86% and 92% by traditional and deep learning method by Nvidia, respectively.