Challenges for Lane Sensing and Tracking Algorithms of the Automated Driving Support System

Lane Recognition is a vital technology for the safe and effective navigations, and it plays a very important role in the development of an self-driving cars. Conventional lane recognition methods have always depended on a mix of manually designed features and lengthy post-processing procedures. Traditional lane detecting systems are made by hand and are therefore specific to particular situations, which increases the likelihood of them malfunctioning on a variety of unpredictable road conditions. The development of Deep Learning (DL) in recent years has completely changed several facets of computer vision, including lane detection. One Kind of deep learning models called convolutional neural networks (CNNs) has shown impressive results in pixel-level tasks like lane segmentation. Without the requirement for human feature extraction, CNNs may automatically extract hierarchical features from the input data. Because of this, they may adjust to a variety of visual patterns and circumstances, possibly providing more robustness than conventional techniques. Some of the common problems are roads without visible lane lines, surfaces damaged by debris or fractures, and hazy or dim lane markers. Under such circumstances, CNN-based models may perform less well, which could result in errors and even safety risks. Moreover, a lot of modern CNN models require a lot of processing resources because to their complex designs. They are not feasible for deployment in embedded systems, which are usually limited by power, size, and processing capabilities, because to this requirement for computational resources. CNN-based lane detection in autonomous vehicles is not widely used due to its need on expensive hardware, especially for applications that need to be integrated into small and energy-efficient platforms. More compact and effective models that can achieve great performance without using a lot of processing power are therefore desperately needed