An Improved Approach for Vision-Based Lane Marking Detection and Tracking

Lane marking detection plays an important role within intelligent vehicles research. The proposed kernel based lane marking detection method, decreases time consuming and improves the detection performance in heavy traffic scenarios. To this end, a horizontal filter is applied to binarize IPM-view images, a cell based blob algorithm is used to eliminate outliers. Starting points of current lane markings in a camera image are estimated as an initialized stage of lane detection. A multi-density kernel based method is introduced to fit quadratic parabolic marking lines. In the end, the detection results are evaluated. Obtained results show that this method is capable of robustly and accurately detecting lane markings in different road and urban traffic scenarios.