Lane detection tecnique for autonomous cars

In this thesis, the aim is to develop and evaluate lane tracking algorithms for autonomous cars. Lane tracking is a critical component for the reliability and efficiency of autonomous driving systems. Lane tracking algorithms are divided into two categories: machine learning-based methods and traditional image processing techniques. Machine learning-based methods work with models trained on specific datasets and may experience performance drops in scenarios outside of these datasets. The success rates of these methods are directly related to the diversity and scope of the dataset used [1]. Traditional methods, on the other hand, operate independently of the dataset and can be implemented quickly as they do not require model training. These methods utilize classic image processing techniques such as edge detection and image segmentation to extract lane information. In this research, the effectiveness of traditional methods for lane tracking has been evaluated. Innovative approaches, such as making the horizontal filter neighbourhood value (τ) adaptive to the brightness of the image, have been employed. The adaptive method facilitates the accurate detection of lanes under varying lighting conditions. Experimental results have shown the positive effects of adaptive τ selection on lane detection. These findings offer significant contributions to enhancing the reliability and efficiency of autonomous driving systems.