Improved Road Marking Detection and Recognition

摘要

A traffic sign is usually represented by graphics or symbols. A driver-less vehicle, moving on a road, is expected to read traffic signs by means of mounted cameras and subsequent video analysis. In our proposed system, we first detect a location of a visual sign from captured video frames and confirm then the candidate area in the frames; then, the captured video frames are transformed by inverse perspective mapping into bird's-eye-view images which are to be classified into text signs or symbols. These road markings are recognised using histogram-of-oriented-gradient features and support vector machines. The proposed method is validated using datasets recorded by Wuhan University, and also by doing actual tests on roads in New Zealand; the recognition rate and the recognition speed are improved compared to previously achieved results.