Research On Building Interactive Extraction Method Based On Oblique Image

UAV tilt photography is a new and emerging technology in the field of remote sensing surveying and mapping.It combines photogrammetry and computer vision technology,and has more applications in 3D modeling of digital cities.With the rapid development of drone remote sensing systems,tilt aerial photography can acquire richer feature information quickly and at multiple angles due to its unique and flexible photography methods,and it can provide better spatial resolution and make images The features,such as the color,texture,and shape of the upper features are more prominent.Therefore,it is possible to use the high-resolution remote sensing images to accurately identify the ground features.However,due to the complex relationship between the ground and remote sensing images,the automatic computer interpretation cannot meet the accuracy requirements.In practical applications,whether in two or three dimensions,most of them rely on manual visual interpretation.,And the efficiency is relatively low,facing a large amount of remote sensing data has shown serious shortcomings.Based on the above considerations,the interactive extraction method is currently a feasible alternative.This method makes full use of human recognition ability and computer processing ability,which not only guarantees the accuracy of understanding translation,but also improves the efficiency of understanding translation.The research in this paper focuses on the interactive extraction method of buildings on remote sensing images and 3D models.The research contents are as follows:(1)In order to solve the problem that Gaussian mixture model needs offline learning and K-means algorithm to perform initial clustering to calculate EM parameters,this paper improves the initial clustering method of Gaussian mixture model accordingly.Firstly,manually select training samples on the image and calculate the EM parameters of the model through manual interaction.Then set appropriate thresholds,traverse the image for binarization,then use mathematical morphology to remove some small patches,and finally pass edge detection.The algorithm extracts the exact contours of the building.The experimental results prove that the method in this paper can accurately,efficiently and stably extract the building contours,with an accuracy of more than 90%.(2)Aiming at the problem of inefficient building collection using three-dimensional models,this paper proposes a method for building extraction based on model cutting.First draw a line on the building model through user interaction,cut the building monomer vertically,and then use the plane to cut the building monomer every corresponding height and repeat the contour lines on each plane.Structure,forming a closed curve.Finally,the maximum and minimum algorithm was used to screen out the appropriate building contours.The building contours extracted by the method in this paper have high accuracy and fast efficiency,so they can be used as DLG data production,which has good practical value.(3)Aiming at the traditional method of collecting building information using a three-dimensional model,it is time-consuming and labor-intensive to adjust the model inflection points and manually connect the contour lines.This paper proposes a method for building extraction based on voting mechanism.First collecting the inflection point does not need to adjust the position of the model deliberately,and then open a suitable image window with this point as the center,then project the triangulation into the image window,extend the line segment and discretize,and use the largest pixel value as the initial inflection point Coordinates,and finally use the point cloud on each side to perform a straight line fitting and right-angled to obtain the final building outline.The contours of buildings extracted by the method in this paper have high accuracy and high efficiency,which greatly reduces the labor intensity of workers.