一种基于消失点和颜色过滤器的车道线检测算法

针对经典霍夫车道线检测方法实用性较差,无法准确区分车道线和路沿与应用道路场景简单等问题,提出 了一种基于消失点和颜色过滤器的车道线检测算法,不仅提高车道线检测的准确率,而且能够应用较复杂行车场 景;首先,对行车视频连续五帧图像进行预处理,获取行车环境下车道线消失点位置,能够自适应选取行车环境图 像的感兴趣区域(Region of Interest,ROI);然后,对 ROI 图像根据车道线颜色特征进行过滤得到二值图像,获取二 值图像中所有连通区域质心和倾斜角等数据,通过结合消失点特征和角度阈值进行限制,筛选记录符合车道线特 征连通区域的数据,接着分割较大区域获取更多质心点,识别漏检符合车道线特征的区域质心点;最后,对获取的 质心点使用最小二乘法进行拟合并标识车道线;实验结果表明:算法能够在多场景道路上快速准确的检测出车道线,与经典霍夫算法进行仿真比较,算法具有一定的鲁棒性和实时性。

A Lane Detection Algorithm Based on Vanishing Point and Color Filter

A lane line detection algorithm based on vanishing points and color filters was proposed to address the problems of poor practicality of the classical Hough lane line detection method its inability to accurately distinguish between lane lines and road edges and the simplicity of the applied road scenes. The proposed algorithm not only improves the accuracy of lane line detection but also enables the application of more complex driving scenarios. First five consecutive frames of the driving video were pre-processed to obtain the location of the vanishing point of the lane lines in the driving environment and the Region of Interest ROI of the driving environment image could be selected adaptively. Then the ROI image was filtered according to the lane line color feature to obtain a binary image and data such as the centroid and tilt angle of all connected areas in the binary image was obtained. By combining the vanishing point feature and angle threshold to limit the data in the connected area conforming to the lane line feature was screened and recorded. The data of the area was then divided into a larger area to obtain more centroid points and the area centroid points that were missed and met the characteristics of the lane line were identified. Finally the least squares method was used to fit the obtained centroid points and identify the lane lines. The experiment shows that the algorithm can quickly and accurately detect lane lines on multi-scene roads. Comparing the classical Hough algorithm the algorithm has certain robustness and real-time performance.