基于融合分割和LASSO回归的实时车道偏离预警

在有路面污染、标识干扰等复杂高速道路环境下,车道偏离预警系统的鲁棒性和实时性会变差。为此提出一种利用两种算法融合分割和LASSO回归模型进行车道线检测和偏离预警的新方法。首先,分别采用TopHat算法和OTSU算法分割出车道线背景并进行"与"运算融合,据此准确提取出车道信息;其次,分两步检测车道线,第一步基于概率Hough变换进行直线检测,将检测出的车道线位置设为动态ROI区域并进行卡尔曼滤波跟踪处理,第二步是基于LASSO多项式回归模型对车道线再次进行参数估计和拟合,以改善使用最小二乘法时的过拟合问题;最后,根据设置的虚拟车道线和角度模型进行车道偏离预警决策。实验结果表明,所提出的方法在复杂道路环境下的平均正检率为96.07%,检测速率可达到32 ms/帧,即具有良好的鲁棒性和实时性。

Real-time lane departure warning based on fusion segmentation and LASSO regression

Robustness and real-time performance of lane departure warning system would deteriorate in the complex highway environment with road surface pollution and traffic sign interference. Therefore a novel method of lane detection and departure warning based on fusion segmentation by two algorithms and LASSO regression model is proposed. Firstly, the lane line background is segmented by using TopHat and OTSU algorithms respectively and then fused by AND operation so that the lane information can be extracted accurately. Secondly, the lane line is detected in two steps. The first step is to use probabilistic Hough transform for straight line detecting. The preliminarily detected lane location is set as the dynamic ROI area which is then tracked by Kalman filter algorithm. The next step is to use LASSO polynomial regression model to estimate and fit the lane line parameters again, so as to solve the problem of over-fitting when using least square method. Finally, the decision of lane departure warning is made based on the virtual lanes and angle model. Experimental results show that the proposed method has an average detection accuracy of 96.07% and an average detection speed of 32 ms/frame in the complex road environment, which can meet the real-time and robust requirements for lane departure warning system.