Optimized generalized hough transform for road marking recognition application

The road markings recognition is an important research in the field of driverless cars. The generalized Hough transform (GHT) is effective for detecting and recognizing contour objects as road markings. While the precision rate of GHT is not very high in applications. This paper presents an edge-type based generalized Hough transform (ETGHT). The edge-type is obtained by multiple thresholds partition of a proposed edge feature and is recorded by multiple R-tables. The edge feature is calculated by a breadth first search strategy using the location and gradient direction of the edge points. In application, a road marking recognition framework based on ETGHT is presented. First, an edge extraction method based on differential excitation is used to obtain the image contours. Then the edge-type feature of the edge points of input image is extracted to determine the corresponding R-table. In the voting stage, a peak region screening processing is used to improve the system's precision rate. Experimental results have shown that the proposed method provides significant improvement of precision rate while ensuring the recall rate.