On average four people get killed and 90 people get seriously injured on Australian roads every day, according to research conducted by the Australian government. Intelligent systems can help to reduce the number of accidents on the roads by creating tools to monitor the driver’s behaviour according to the lane path. Lane detectors are important tools and many approaches have been applied to solve the lane detection problem, but most of them are not capable of dealing with dynamic lighting efficiently. The current approaches still rely in expensive computational effort and long training phases. The reliability of the lane detection is a key component for safe navigation on the roads and an automatic solution becomes an interesting topic to explore. This research proposes a vision-based software architecture which combines an appearance-based analysis with salient point tracking, automatically adapting to different lighting situations. It aims to increase the efficiency and accuracy for the detection of the left- or rightmost lane boundary (ego-lane) providing a flexible software architecture to support drivers and to be extended to driverless vehicle control. The experiments exposed a series of important facts to address. First, the adaptive appearance-based analysis is capable of segmenting multiple lane painting marks. Second, the system can reduce the false positive detection by using an efficient road model. The proposed model guides the image processing cues to focus on relevant Regions-Of-Interest (ROIs) for lane marking detection instead of the analysing the whole image.
Unless otherwise indicated, works by Griffith University Scholars are © Griffith University. For further details please refer to the University Intellectual Property Policy.