Research Of Highway Rainy Weather Detection Based On Deep Semantic Segmentation

With the rapid development of our country's economy,China's transportation industry is facing significant challenges as it welcomes tremendous development.As a link between domestic cities,highways will ensure that the safety and smoothness of the road will greatly affect the economy and life of the city.Therefore,the study of highway conditions under bad weather will have significant implications for highway safety.The rainy weather conditions in bad weather occur frequently in many seasons across the country,and the frequency and the scope of influence are relatively large.Therefore,the purpose of this thesis is to achieve detecting rainy weather by processing the video data of highway surveillance.This thesis makes full use of the highway surveillance data information,and uses the state-of-the-art deep learning based semantic segmentation method to achieve the purpose of detecting rainy weather through video data.Through the extraction and analysis of the highway surveillance video data,the algorithm can comprehensively determine whether the weather condition of the road section is rainy from the reflective detection of the road surface and the wet skid detection.This thesis first studies the semantic segmentation algorithm based on deep learning,and compares the performance and accuracy of the semantic segmentation network on a self-made highway rainy weather detection data set.It also proposes a combination of highway monitoring video reflection and road wetness detection.The multi-task model tests the performance of the model on the rainy day classification problem;this thesis then proposes a comprehensive formula for the determination of rainy weather based on the unified formula of road reflection and road slippery.The experiment shows that the highway rainy weather detection algorithm combined with the judgment formula would achieve the purpose of rainy weather detection through the highway suiveillance.Finally,the article discusses the improvement of the detection algorithm for highway rainy weather and improves the accuracy and training speed of the algorithm.