Driver assistance functions on marked innercity intersections require a reliable detection of intersection lanes. Due to the high complexity of intersection scenarios and the large amount of clutter that is usually encountered in urban areas, existing highway lane detectors are not applicable for this task. In order to detect the lanes on marked innercity intersections, we adopt a two step process of first detecting individual lane marking segments and then aggregating these segments to appropriate lanes. We assume that a lane detector for innercity scenarios should be capable of handling lanes with arbitrary orientation and curvature and thus may not rely on simple geometric models. In order to achieve this, in the paper we first derive a set of features from the lane marking segment data. These features are fed into a support vector machine. The support vector machine determines whether the lane marking segments belong to the same lane or not.