Research On Image Semantic Segmentation Algorithm Of Complex Traffic Scene

As the key link of autonomous driving,environmental perception is the guarantee of driving safety and intelligence.In environment perception,an accurate understanding of traffic scene semantic information is the basis for safe driving and the key issue of current autonomous driving technology.Therefore,it is of great practical significance to study a high-performance image semantic segmentation algorithm.In this thesis,the image semantic segmentation technology under complex traffic scenes is studied,and the complex traffic scene segmentation model aimed at improving image segmentation quality is constructed by means of deep learning,and a multi-scale fusion image semantic segmentation algorithm combining superpixel segmentation and converse neural networks is proposed.The specific work is as follows:The evaluation system of semantic segmentation model is constructed.In order to explore the principle of semantic segmentation based on deep learning and evaluate the semantic segmentation model scientifically,this thesis analyzes the basic working principle of convolutional neural network,combs the current semantic segmentation algorithm,selects the common evaluation index of semantic segmentation,and determines PA combined with MIoU as the evaluation system of this model.The semantic segmentation data set of complex traffic scene image is constructed.In order to obtain the data set suitable for this thesis,this thesis analyzes the current mainstream public data set Cityscapes,and extracts the applicable image data and its label files in the Cityscapes data set.The image data of ice and snow weather is collected and manually labeled by labelme to expand the traffic scene data under special weather conditions.These two data sets are fused as the data sets of this thesis.The multi-scale feature fusion image semantic segmentation network is constructed.Based on PSPNet network model,the void pyramid pooling module istransformed.The method of extracting the same feature map from multiple branches of the module is changed into the form of extracting multi-level feature map from multiple branches to form multilevel feature fusion module.Therefore,the parallel extraction and fusion of multi-scale feature information is realized.The multi-scale feature fusion image semantic segmentation network is optimized.In order to solve the fuzzy edge problem of scale feature fusion network segmentation results,the super-pixel segmentation module is added,extracted the low-order features of the image by SLIC(simple linear iterative clustering)algorithm and improved the semantic segmentation network segmentation accuracy of multi-scale feature fusion image by combining super-pixels with convolutional neural networks.