Automatic Recognition And Detection Of Road Traffic Elements From Drone Image

Road traffic elements are an important part of roads,including road centerlines,road intersections,zebra crossings,bus stations,and roadside parking spaces,etc.Accurate recognition and detection of these elements can provide important data support for autonomous driving,improving intelligent transportation systems,promoting smart cities,and updating basic traffic geographic information databases.However,there are still some problems with the extraction of traffic elements: 1)There is a lack of UAV road traffic multi-element datasets.Research on road traffic element information is rare,and there is a lack of publicly available research datasets.2)The detection performance of multi-scale and juxtaposed dense road traffic elements is poor.The scale difference of road traffic elements is large,and there are difficulties in detecting multi-scale and juxtaposed dense elements in complex scenes,with low detection accuracy.3)The small object elements in the multi-scale road traffic elements are easy to miss.The characteristics of small object road traffic elements are easy to be ignored,especially when they are occluded.Due to the limitations of traditional vehicle-mounted cameras,only a small part of the road element information can be obtained,which is not conducive to the acquisition of large-scale road traffic elements.UAV imagery data has the advantages of convenience to obtain,fast data acquisition,clear imaging,and high image resolution,providing favorable conditions for the acquisition of large-scale road traffic elements.Therefore,this paper proposes a method for automatic recognition and detection of road traffic elements based on UAV imagery,and carries out related research on the automatic recognition and detection of road traffic multiple elements.The main research of this paper is as follows.(1)We propose an improved YOLOv4 network algorithm combined with Efficient Channel Attention(ECA)to automatically recognize and detect road traffic multiple elements in UAV imagery.The method focuses the network on feature information by using ECA.Additionally,the CIo U loss function is used to consider the geometric relationship between the object and the test object,solving the overlapping problem of juxtaposed dense test element anchor boxes and reducing the rate of missed detection.The mean Average Precision(m AP)of this method is 90.45%,which is 15.80% better than the average precision of the original YOLOv4 network.This method provides a new approach for updating basic traffic geographic information databases.(2)An adaptive spatial feature fusion(ASFF)YOLOv5 network(ASFF-YOLOv5)is proposed for the automatic recognition and detection of road traffic multiple elements.ASFF-YOLOv5 proposes an ASFF strategy that incorporates the Receptive Field Block(RFB),which improves the scalability of features and enhances the detection performance of small objects.According to experimental results,the m AP score of this method is 93.1%,which is a significant improvement of 19.2% over the original YOLOv5 model.Using this method can solve the problem of wrong or missed detection of road traffic multiple elements,enhance the accuracy of detection of road traffic multiple elements,and achieve automatic detection of road traffic multiple elements.