Method Of Extracting Floor In 3D Model Of Oblique Photogrammetry Based On Fourier Transform

Oblique image-based modeling is one of the most important methods for 3D reconstruction of urban area.Compared with the traditional vertical aerial photography,the laser scanning and close-range photography,oblique photogrammetry acquire the texture of the building facade,the cost is lower and high degree of automation modeling.Therefore,oblique images are widely used in digital 3D city modeling,urban management and planning,and other fields.However,the tilt image model is continuous,and the model object cannot be independently selected,searched,edited,etc.,so it's need to monomerization the model object.The commonly used monomerization software includes DPModeler and CityBuilder,etc.DPModeler is earlier used to achieve the real-time three-dimensional reconstruction,the feature is the large degree of human participation,can achieve model physics monomerization and various fine processing;CityBuilder is Terry Skyline's products,a high degree of automation,can achieve logical single model and various editing operations.With the continuous advancement of smart city construction,the city's digitization has become increasingly demanding.In the management and service of smart cities,such as real estate registration and public security "Three Realities",we need not only the information of individual buildings,but also the location and attributes of building floors,which requires the monomerization of the urban building models and a further refinement of the object.Layered monomerization is one of building refinement methods.At present,the monomerization is mainly based on the laser scanning point cloud which can capture the fine structure of the object:firstly,structural point cloud of floor is extracted manually and is used to fit the structure,then,the building layered monomerization is complete based on the regular changes of floor in the vertical direction.Drawbacks of this method is that it only considers the characteristics of local floor and assumes that floor changes with the same interval height.These drawbacks are limiting the scope of the method's application.This paper proposes a method based on Fourier transform to monomerize the 3D model of the building.The Fourier transform transforms the time-domain signal to the frequency domain,which is characterized by analyzing the characteristics of the frequency domain,obtaining the regular characteristics of the signal that do not change significantly in the time domain,and simulating some characteristic changes with specific form of signal.The building features includes windows,balconies,and air conditioners.In the oblique image model,these features are irregular in texture and structure,in addition,the oblique image may be obstructed,resulting in distortion of the local texture features,making it difficult to accurately extract the model surface features.However,from the overall analysis of the buildings,these features showed regular changes with building floors changed.Combined the building model texture features and the Fourier transform characteristics,the characteristics of the model changes can be extracted and a further layered location of building is achieved finally.Layering of building model in this paper is mainly divided into two parts.In the first step,this paper combined the idea of dimensionality reduction to orthographically project the texture of a three-dimensional model to a two-dimensional plane to reduce the complexity of data processing.Taking into account the building floor change law in the vertical direction,so the projected texture image is organized into three kinds of column vector signals,gradient signals,cumulative signals and integrated signals,discuss the advantages and disadvantages of the three signals separately,the conclusion is that the integrated signal combine the front two signal's advantage and can better reflect the building texture changes.In the second step,a detailed analysis of the Fourier transform to achieve a layered approach to various texture signals.In order to make the Fourier transform original signal in the same order of magnitude,we first need to block the texture image.For the regularized texture image,we use the Fourier transform to extract the layer height,obtain the cosine function corresponding to the layer height,and solve the cosine function pole,for the small value,the initial layered position of the texture image is obtained,adjust the cosine function period,and search the symmetry line of the corresponding layer height region.According to the type of layer height,divided into single-valued layer height and multi-valued layer height.The layering method of two data types are combined with the Fourier transform amplitude spectrum map,the corresponding height of the cosine function and symmetry Line,achieve accurate layering,buffer analysis between different modules,achieve different levels of hierarchical fusion module.Finally,the applicability of layering method is validated by building model data with different structural complexity and different texture complexity.According to the texture image layering position and orthographic projection relationship,solve the hierarchical position of model and display the model layer.