Semantic Segmentation Method of Urban Traffic Images Based on Internet of Things

In response to the problem of the imbalance in the weights of different categories in the recognition of complex scenarios in traffic roads, the paper proposes an urban traffic ISS (Image Semantic Segmentation method) based on the Internet of Things. The approach is to use Full-CNN (Full Convolutional Neural Network) for urban traffic image feature extraction, and extend DenseNet architecture to FCNDN (Full-CNN-DenseNet) to carry out, with the help of DenseNet characteristics to reduce the number of parameters, in order to achieve automatic processing and intelligent analysis of urban traffic information. In the semantics of urban landscape data sets, our method has higher segmentation accuracy and processing speed in traffic signs and vehicle recognition. Specifically, FCNDN is 10% more accurate than most networks such as eNet and ContextNet, and can achieve real-time speeds of more than 30 FPS. In the PA index performance comparison, the FCNDN model can achieve a pixel accuracy of 93.7%, while the mRecall model can achieve a pixel accuracy of 89.5%, both of which show good advantages in all models. Ultimately, functions such as real-time urban traffic condition monitoring, rapid accident handling, and enhanced environmental perception can be realized, which is of significance for urban management departments in terms of traffic planning and optimization, rational allocation of resources, and support for the development of green transportation.