基于计算机视觉的高速路面状态检测方法

Computer Vision Based High-speed Pavement Condition Detection Method

路面状态检测技术是辅助高速公路监管部门及时发现结冰、积雪等不良路况的重要手段。针对传统基于硬件设备的路面状态检测技术存在成本较高、检测范围受局限等问题,提出了一种基于计算机视觉的高速路面状态检测方法。该方法首先将融合了空间注意力机制和通道注意力机制的Attention模块与具有高分割精度的U-Net网络相结合,对路面区域图像进行分割提取;之后实现了一种基于循环生成对抗网络的路面阴影消除算法,对已经提取的路面图像进行阴影消除;最后基于残差结构构建了路面状态分类器,实现高准确率的路面状态检测。实验结果表明,采用该方法可以消除路面状态检测过程中高速公路两侧景物和路面阴影对状态检测带来的干扰,实现了对路面干燥、积雪、积水和结冰4类状态的准确识别。基于高速公路监控视频进行测试时,识别率可达97.9%。

Pavement condition detection technology is an important tool to assist highway supervisory departments in detection of icy,snowy,and other undesirable pavement conditions timely.A computer vision-based method for highway pavement condition detection is proposed.In this method,Attention module with spatial attention mechanism and channel attention mechanism is combined with U-Net network with high segmentation accuracy first.Then,a recurrent generative adversarial network-based pavement shadow removal algorithm is implemented to remove shadows from the already extracted pavement image.Finally,a pavement condition classifier is constructed based on the residual structure to achieve high accuracy of pavement condition detection.The experiment results show that the interference caused by the scenery on both sides of the highway and the shadows on the pavement can be eliminated during the process of pavement condition detection when the proposed method is adopted.The accurate recognition of four types of pavement condition including dry,snowy,waterlogged and icy is achieved.When testing based on highway monitoring video,the recognition rate can reach 97.9%.