| Car refueling is a common life scene.In order to solve the problem of unsanitary and inconvenient manual refueling,automatic refueling has become a new development trend.However,there are still many problems and challenges in the realization of automatic refueling,among which the perception of fuel tank cap is a key step to realize automatic refueling.This paper studies the image segmentation and localization of the fuel tank cap in order to provide accurate spatial position information of the fuel tank cap for the realization of automatic refueling.The premise of fuel tank cap localization based on binocular vision is that the fuel tank cap area can be accurately extracted from binocular camera images.Considering the effectiveness of the algorithm and the requirement of automatic refueling scene,a lightweight fuel tank cap image segmentation network FTC-Net based on global attention is proposed in this paper,which not only reduces the number of parameters and computation,but also fully learns the spatial position information of the image,and improves the segmentation performance of the fuel tank cap image.In order to solve the problem of lack of special dataset for fuel tank cap image segmentation,a dataset containing 13920 fuel tank cap images is established,which can support the training of a robust fuel tank cap image segmentation network.By using this dataset to train and test the FTC-Net and other three popular image segmentation models,and analyze the size of the four models,it is proved that FTC-Net can accurately achieve fuel tank cap image segmentation and meet the needs of lightweight.Considering the localization requirements of automatic refueling scene and the scarcity of computing resources,two projection models of spatial plane graphics are studied and analyzed.And a stereo matching method for finding the geometric center of connected domain based on first-order central moment is proposed.The stereo matching process in binocular vision localization of fuel tank cap is simplified.This method is used to locate the geometric center of the automobile fuel tank cap in threedimensional space,and the experimental results show that this method can accurately locate the geometric center of the fuel tank cover.Considering the relationship between the perception system and other systems in the automatic refueling scene,a fuel tank cap perception system is designed.By encapsulating and embedding the model,method and other basic functions proposed in this paper,the image segmentation and binocular vision localization of the fuel tank cap are realized in the form of client,which provides convenient and reliable sensing system support for the realization of other systems in the automatic refueling scene.This paper includes 44 images,20 tables and 66 reference doucumentations. |
