Abstract
High accident rates in urban environments are caused by lane change manoeuvres, especially in the surrounding of intersections. For highly automated driving in the urban area, it is essential to know the precise position of the vehicle. Furthermore, it is important to understand the surrounding context in complex situations, e.g. multilane crossings and turn lanes. To understand those scenarios there is not only the task to detect the lane markings, but to detect the painted arrows inside the lanes.
The proposed paper presents two different approaches for the detection and classification of such lane marking arrows using fisheye cameras. The methods have in common at a first stage that, the captured image is transformed to a bird view. The classification is split into two different ways, which the results are evaluated and validated.
The first proposed method is based on the extraction of contour, the collection of geometrical features and the generation of the Fourier coefficients from the transformed image which are used as input for the Support Vector Machine for the classification. The usage of Fourier coefficients is well known by OCR problems and it is evaluated for the recognition of lane marking recognition in this proposed work.
The second one use a form-classifying HOG (histogram of oriented gradients) cascade to generate marking arrow hypotheses and classify them in different classes based on the marking direction painted on the street.
The proposed method shows that a high detection confidence is achieved, proofed with validation datasets and in practical usage.
KEYWORDS Fisheye camera, arrow detection, arrow classification, vehicle localisation