摘要
Abstract
As the total mileage of highways in our country increases in a rapid pace,the number of traffic accidents caused by agglomerate fog also grows continuously.Due to its characteristics of sudden formation,small coverage,and high randomness of dis-tribution,agglomerate fog is difficult to be detected.In the situation of high concentration,agglomerate fog would pose serious im-pact on road safety.Traditional fog detection methods in general require wireless sensors or lasers to build surveillance stations,suf-fering from the disadvantages of complicated technical flow,implementation difficulty,low economic efficiency.To address the above-mentioned limitation,this paper proposes a deep learning-based highway agglomerate fog detection method that relies on highway monitoring to achieve fast detection of agglomerate fog level and significant reduction of fog detection cost.With road moni-toring images as the input,the proposed detection method starts with obtaining the binary features of lane lines through the lane line segmentation network and extracting dense fog features by designing the dense fog area segmentation network branch.Then,the de-tection method integrates the visible lane line features,the dense fog area features,and the road gradient features into a feature fu-sion network to train the horizontal visible road distance,and in turn predicts the agglomerate fog level on the road.Experimental re-sults demonstrate that,the fast agglomerate fog detection method proposed in this paper is cupable of predicting the agglomerate fog level based on road monitoring images,in terms of high accuracy and robustness in complex foggy road environments.关键词
团雾检测/深度学习/车道线检测/融合特征Key words
agglomerate fog detection/deep learning/lane detection/fusion features分类
信息技术与安全科学
