Optimal Scheduling Method Of 3D Navigation Scene Model Based On Visual Saliency

In recent years,the application of Geographic Information System(GIS)has not only improved the ability to process and analyze geospatial data,but also promoted the development of navigation technology.On the one hand,although the current twodimensional map navigation can meet the simple navigation guidance needs of people in daily life,with the prosperity and development of urban transportation,viaducts and ring roads rise on the ground.When facing complex urban scenes,Two-dimensional real-time navigation often fails to meet people's needs for precise navigation,and it is difficult to ensure a full grasp and understanding of the surrounding environment,which brings inconvenience to daily travel.On the other hand,in addition to assisted navigation by manned vehicles,cities based on cameras,Li DAR,Global Navigation Satellite System(GNSS)and other devices to obtain the vehicle's position,obstacles and distance in front,and surrounding scenery Unmanned navigation in complex scenes is also driving the vigorous development of 3D real-time navigation technology and its applications.Regardless of the application scenario,most of the commercial software on the market is still at the level of 2D real-time navigation.There are still many key issues waiting to be solved for how to cross from 2D to 3D.The huge amount of 3D city scene data cannot be loaded all at once.It is necessary to schedule and hierarchically display the scene data according to the viewpoint parameters,so as to realize the goal of real-time calling the front 3D scene model consistent with the driving direction and position according to the user's driving track.In order to solve the problem of optimizing the scheduling method of the 3D navigation scene model based on visual salience,this paper studies and optimizes the visual saliency algorithm,designs a method to obtain spatial coordinate information and shooting posture information from real images,and proposes a visual saliency-based The hierarchical scheduling and visualization method of multi-levels of detail(LOD)model of the threedimensional scene makes the display results more in line with the human visual attention mechanism.The main research contents are as follows::(1)Analyze the domestic and international research status of computer vision development,visual saliency measurement algorithm,image positioning and posture information analysis,and 3D scene LOD model scheduling strategy,and summarize the existing problems in current research.Contrast and analyze different visual saliency algorithms,analysis methods of image positioning and posture information,and scheduling methods of 3D scene LOD models,and summarize them.(2)Optimize the Minimum Barrier Distance(MBD)algorithm and the background prior algorithm,and compare and summarize the algorithm results of different features images.The optimized algorithm is based on the Fast raster-scan algorithm.Iterative method is used to obtain the approximate MBD value.The background boundary information map is used to enhance the brightness and normalize the processing.The optimized background prior algorithm uses super The shortest distance between pixel blocks,measure their similarity,quantify and calculate the concept of boundary connectivity,then use the background contrast weight of superpixels to enhance the contrast of the result,and finally perform the least squares calculation to optimize the display effect.(3)Explore the analysis methods of positioning and posture information based on visual sensors(that is,position information and shooting posture information),and use the resolved geographical space coordinates to locate and fix the angle of view and Lay the foundation for the scope of sight.This article uses the images taken by the Android system mobile phone as the experimental data source to analyze and obtain the geographic location coordinates and shooting attitude information(including pitch angle,azimuth angle,and flip angle)at the time of shooting.(4)A data organization method based on a pyramid structure and an index mechanism based on a quadtree are established to evaluate the visual saliency of different LOD hierarchical models,and determine the lowest LOD hierarchical model scheduled under the premise of ensuring visual saliency accuracy.The three-dimensional scene is located based on the spatial coordinate information parsed in the foregoing,the viewing angle of the threedimensional scene is fixed based on the resolved shooting posture information,and the threedimensional scene LOD model scheduling method is based on the visual saliency of the captured real-time image as the criterion.Among them,the regions with relatively high visual saliency call the higher level of detail level model,and the relatively low regions with relatively low level of detail call the relatively low level of detail level model.Then use the actual shooting image data to simulate the data scheduling and display during threedimensional navigation,compare and analyze the original scheduling method and the scheduling method discussed in this paper,and evaluate the difference between the display effect and the real-time response of the two.