Research On Vector Reference Map Preparation And Matching Positioning

Scene matching navigation system is an autonomous navigation system which does not rely on satellite navigation.It has the characteristics of strong anti-interference and high positioning accuracy.The preparation method of reference map is an important part of the scene matching technology and is closely combined with the matching algorithm,which determines the performance of navigation.Raster reference map has many problems,such as large amount of data,high cost of acquisition and limited coverage,while vector data has the characteristics of flexible viewpoint,lossless transformation and small amount of data.According to this,a method of using vector data to prepare a vector reference map in real time is proposed,and the method of matching and selecting the suitable-matching area based on the vector reference map are proposed.The main work is as follows:In order to ensure that the aircraft can prepare vector reference maps in real time during the flight,vector data reading and fusion with large calculation requirements are completed in advance.For application scenarios where the viewpoint is moving and the scene is fixed,the BSP tree is used to manage the vector data and load it on the aircraft in advance.During the flight,a vector reference map is flexibly prepared based on the track information and camera parameters in real time.Experiments show that this method can accurately prepare vector reference maps in different application scenarios.The average mapping time is linearly related to the scene complexity and independent of the flight altitude and attitude of the aircraft,and can meet real-time requirements.In order to solve the problem that the vector reference map and the real-time image cannot be directly matched due to huge differences,an image conversion method based on CGAN is proposed.The loss function based on the edge distribution is used to make the real-time image and the vector reference map have consistent edges,and the real-time image is converted to the vector-style image to reduce the difference to complete the matching.Experiments show that this method can effectively convert real-time images into vector-style images and has good generalization ability.It has higher matching accuracy and matching success rate in different flight tasks.Aiming at the problem that the method of selecting the suitable-matching area of raster reference map is not suitable for vector reference map,a method of selecting the suitable-matching area based on GBDT model is proposed.Firstly,the principle of richness,uniqueness and stability of raster reference map is analyzed,and the characteristic indexes describing the distribution and richness of vector elements as a whole and the characteristic indexes describing the distribution,richness and repetitive structure of line and surface elements separately are designed.Then,the regression model between the suitable-matching features and the matching probability is constructed,and the suitable-matching area is selected according to the prediction results of the model.Experiments show that the method can select the suitable-matching area with high recall and accuracy.