Estratégia para detecção e rastreamento de faixas rodoviárias utilizando uma câmera monocular

Abstract

In the automotive field, driver assistance systems are developed to reduce the collateral effects of the actual level of mobility, such as traffic jams and accidents. In a future perspective, it is intended to achieve the level of autonomous and cooperative driving based on sensor networks. The proposed strategy for road lanes detection and tracking fits as a functional requirement for some of these systems, as the Lane Departure Warning and Lane Keep Assist. The algorithm development was structured based in three processing levels: low, mid and high levels. The low-level processing enhances the input image, the mid-level processing is an interest feature extractor, and the high-level is the lane position tracking strategy. The algorithm's response evaluation, for the chosen samples, was realized with metrics based on the deviation between the tracked and the original lane. The strategy shows good accuracy levels at the ideal scenario, including shadows, curves, and road slope. However, the accuracy is impaired if the lane is dashed, badly signalized and if road reflection or dazzle harm the image capture by the camera.