Learning of Kalman Filter Parameters for Lane Detection

This paper presents a framework for learning of system parameters for vision-based lane detection systems. Learning is achieved by ground-truth data based optimization of a performance measure evaluated on video sequences. Different options for evaluating the performance of lane detection systems are discussed, and in order to allow for a linear combination, we show how these performance measures can be normalized. The approach presented is applied to the optimization of the state noise variances of a Kalman filter. The surroundings around the located solutions are examined by 2D-grid analysis. It turns out that this approach leads to the same regions for robust parametrizations independent on the starting conditions for the optimization, and thereby a well generalizing parameter set can be obtained