Abstract：To solve the problem that the uncertain external environment was easy to interfere the existing lane detection algorithms and lead to false detection, a new lane detection method was proposed based on improved Hough transform. The image was filtered, and the edge was recognized to perform the two valued preprocessing. After preprocessing, the feature points containing lane edges were extracted by behavior units. The feature points were constructed according to Euclidean distance, and the relationship was established at longitudinal direction. The clustering algorithm was used to combinate the feature points and isolate the feature points with clear targets, and the calculation error of feature point interference and Hough transform was reduced to improve realtime performance and calculation accuracy. The algorithm was checked by lane images with about 4 000 frames of different time periods and external environment. The experimental results show that the proposed method can well realize lane detection under various environment, and the correct rate is 99.18% in fine weather with that of 97.45% in adverse weather.
2017, Vol. 38 
