Research On Road Scene Semantic Segmentation Technology In Severe Weather Conditions

Semantic segmentation in severe weather road scenarios is of great significance in autonomous driving,highway navigation systems,and many other safety-related applications,but it has not been systematically studied.In general,the image of road scenes in severe weather will increase the difficulty of semantic segmentation and often lead to the decline of semantic segmentation accuracy.Therefore,the performance of the semantic segmentation algorithm on clear images can be regarded as the upper bound of its performance on the severe weather road scene images.If there is no sensing system in autonomous vehicles that can handle such a huge visual change,it will quickly endanger the people around it.Therefore,it is necessary to study the key technologies of semantic segmentation of severe weather road scenes to promote the practical application of autonomous driving systems.The main contents of this article are(1)Aiming at the problem of unsatisfactory semantic segmentation of road scenes under complex and changing weather conditions,it is proposed to first use image enhancement algorithms to restore the images of bad weather road scenes to normal weather.A twostage semantic segmentation method for road scene images and semantic segmentation of their images;road scene images of different severe weather are collected and synthesized.(2)A single image enhancement algorithm that combines boundary perception and knowledge distillation is proposed,and a deep convolutional neural network structure of knowledge distillation + boundary perception is constructed,and the two issues of image boundary extraction and image enhancement are unified in one optimization box.By solving these two key problems,the respective optimization results can be promoted.Based on the bilateral structure,it is proposed to use the space module and the perception module to obtain rich spatial information and large receptive fields respectively.After these two modules,this article introduces a feature fusion module to effectively extract the two modules.The features are combined to get the final semantic segmentation result.(3)Experimental tests were performed according to the proposed method,and the results of image enhancement and semantic segmentation were analyzed and evaluated.The experimental results show that in this paper,using the method of knowledge distillation,the computing time of the image enhancement method is greatly reduced,and the results are close to the complex method;it shows a good segmentation effect for road scenes.