摘要
针对结构化道路的特点,提出一种实用的基于组合模型的车道线自动识别方法。近视场区域采用Hough变换初始检测车道线,远视场区域采用三次曲线模型拟合车道线。车道跟踪用Kalman预测参数动态建立ROI(region of interest),用扫描线法搜索车道线边界点,在车道线间断处用Kalman预测器定位车道线边界,还设计了一个失效判别模块来验证跟踪结果。最后将投影图中检测到的车道线进行逆投影重建,得到实际路面的车道线。实验结果表明,对于不同的车道线种类和在部分车道线被前方车辆遮挡的条件下,该算法均具有较高的实时性和鲁棒性。
A practical method of lane recognition for structural road is proposed based on combined model. Lane is initially detected by Hough transform in near area and curve fitting in far area. In lane tracking, the region of interest （ROI） is established with parameters predicted through the Kalman predictor, the lane boundary points are searched by scanning line method in ROI, and the lane boundary in discontinuous area is located by Kalman predictor. A discrimination module responsible for estimating tracking results is introduced. Finally, a real lane is obtained using inverse projective reconstruction of the lane on the image. For various kinds of lanes and most lanes covered with vehicles ahead, experiment results show that the algorithm has good robustness and efficiency.
出处
《计算机工程与设计》
CSCD
北大核心
2009年第6期1548-1551,1554,共5页
Computer Engineering and Design
基金
北京市教委和北京市基金委重点共同资助基金项目(KZ20041000501)
