Algorithms And Implementations For Monitoring Road Meteorological Condition Based On Artificial Intelligence And Image Recognition

Unfavorable weather conditions have a direct impact on road traffic safety,especially,the main cause of traffic accidents is slippery road caused by rain and snow.Some of the current road weather monitoring methods are inefficient,the rest methods are too costly to apply in large scale.Based on the image recognition algorithm,this thesis designs a method which can obtain the real-time meteorological information of the road surface and classify the road surface condition.In this thesis,two road meteorology classification algorithms based on image recognition are designed,while the road meteorological condition monitoring software is written with C#.Following functions are designed in the software,controlling remote monitoring node,displaying real-time data of monitoring node,querying historical data of monitoring node,etc.The following works are carried out in this thesis:1.A data augmentation method of road surface image is designed using the image style migration network MUNIT.The training data set of MUNIT network is made by frame of the video taken by dash cam,“bag of words” model and SVM training data sets come from MUNIT network.Experimental result shows that the classification accuracy of this method reaches 87.07%.2.The characteristics of the road surface meteorological state images are analyzed,then,the right of the high-level semantics of the road surface meteorology state image is improved heavily in classification algorithm.Dense Net121 is offered to the base frame,and the first two dense layers of Dense Net121 are replaced by residual layers,reducing 2.3 million network parameters without reducing the classification accuracy,the compression rate of the model is 32.86%.3."monitoring node-ECS-PC software" is designed as the communication method.In the ECS,multiple tables are built to store the meteorological data collected by the monitoring node and realize the remote controlling of the node.Meanwhile,the ECS is also responsible for running the classification algorithm of the road meteorological state image and writing the classification results into the database,this method reduces 2.46 MB of memory usage and 14.29% of CPU exclusive time of PC monitoring software per hour.4.The road surface meteorological condition monitoring software is written with C# in the.NET Framework of Windows platform.In the function test,the monitoring software realizes these functions including real-time receiving and displaying multiple meteorological data,remote controlling of nodes,querying and exporting of historical data of nodes,etc.In the stability test,the software memory consumption is stable within 122 MB,and the CPU utilization rate is 36% in saturated operation,overall,the monitoring software and the recognition algorithm deployed in the ECS can work together.