Detection Of Pavement Distress And Foreign Objects Based On Artificial Intelligent Technology

The rapid detection and response of pavement distress and foreign objects is the critical part of daily inspection for highway,which ensures the traffic safety of roads.Detecting pavement distress and repairing them in time that make traffic safety and comfort;timely detection and treatment of pavement throws（abandoned objects）reduce the risk of traffic accidents.In currently,the road scene image（distress,foreign objects,Guardrail,etc.）is easily and automatically collected,but the detection of pavement distress and foreign objects is still based on manually labeling method.The widely used pavement distress detection methods is based on image recognition technology and majorly focuses on pavement cracks detection.Indeed,the accuracy of these methods needs to be improved.The current researches of anomaly detection is mainly carried out under a simple background.Rare researches focus on unknown foreign objects detection on road scene.In order to achieve automated and informatized road inspections,this paper proposes series methods for pavement distress and foreign objects detection based on AI technology.The research works includes:（1）Use Roadway inspection binocular system to collect massive road foreground image data,and on this basis,combine image processing and image enhancement technology to delimit the target detection area,and build a model training library for automatic identification of road distress and spattering objects.（2）Propose a novel Mask R-CNN（AFFM）pavement distress automatic detection model.In the feature extraction stage,the attention mechanism module（Convolutional Block Attention Module,CBAM）and the feature fusion module（Feature Fusion Module,FFM）are combined Body AFFM.（3）Propose an automatic detection method for road foreign objects based on the Center Net model,and combine the image inverse perspective transformation to remove the interference of the sky,signs and signs and other complex traffic environment backgrounds to improve the detection accuracy.The road distress detection method proposed in this paper not only improves the feature expression ability of the model,but also solves the problem of information loss in the feature extraction process of the deep convolutional neural network,thereby improving the detection and segmentation performance of the model;the road surface proposed in this paper The foreign objects detection method regresses to the size of the prediction frame by predicting the key point position of the target to avoid the problem of poor detection effect on road foreign objects of different shapes caused by the constraint of the a priori frame aspect ratio parameter.The experimental results show that the Mask R-CNN（AFFM）model using Res Net101-AFFM-FPN as the feature extraction backbone network has a m AP of 69.73%for foreground image road distress detection;the AP₅₀of Center Net-based road foreign objects detection is70.9%.As mentioned in this paper,the methods for detecting road distress and foreign objects based on Mask R-CNN（AFFM）and Center Net model both have good detection results,and also provides a new working mode for daily road inspection work,which can improve the efficiency of daily inspection work of road maintenance department,and reduce the cost of management,and at the same time provides technical support for building an information-based road maintenance system,which has certain research significance and application value.