Building three-dimensional damage detection method

Abstract

The invention discloses a building three-dimensional damage detection method, which comprises the following steps of: 1, carrying out image acquisition on a target building by using an unmanned aerial vehicle carried with a visible light imager to obtain a plurality of building images covering the whole target building, and generating a building DSM (Digital Subscriber Model); 2, acquiring a top elevation image of the target building; 3, detecting the elevation change of the target building; 4, detecting the top damage of the target building; 5, detecting side damage of the target building; and 6, visualizing the damage of the target building. The method is simple in steps and reasonable in design, the unmanned aerial vehicle is used for shooting the images of the target building at multiple angles, detection of elevation change, top damage and side damage of the target building is achieved, and overall visualization of damage of the target building is achieved.

Description

Technical Field

[0001] The invention belongs to the technical field of image recognition, and in particular relates to a three-dimensional damage detection method for a building. Background Art

[0002] Change detection based on multi-temporal remote sensing images is to extract the changes of objects in different time phases by comparing and analyzing two or more remote sensing images of the same area at different time phases. With the importance of building change detection in urban development monitoring and damage assessment, building change detection technology based on multi-temporal remote sensing images has broad application prospects in civil and military fields. There are many building change detection methods, and they are applied to solve various problems.

[0003] However, the current building damage detection methods are mainly focused on two-dimensional change detection, which mainly uses the grayscale difference between remote sensing images in different time phases to detect the damage information of the top of the building, or the damage information of the facade of the building. The main problem with this type of method is that it can only detect the changes in the top or side of the building in different time phases, and cannot obtain comprehensive three-dimensional damage information. Therefore, its application in building disaster damage detection and impact effect evaluation is limited.

[0004] Therefore, a rationally designed three-dimensional building damage detection method is needed, which uses drones to capture images of target buildings from multiple angles to detect elevation changes, top damage and side damage of target buildings, and realize overall visualization of target building damage. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a three-dimensional damage detection method for buildings in view of the deficiencies in the above-mentioned prior art. The method has simple steps and reasonable design. It utilizes drones to capture images of target buildings from multiple angles to detect elevation changes, top damage and side damage of the target buildings, thereby achieving overall visualization of the damage to the target buildings.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a three-dimensional building damage detection method, characterized in that the method comprises the following steps: Step 1: Use a drone equipped with a visible light imager to collect images of the target building, obtain multiple building images covering the entire target building, and generate a building DSM model; Step 2: Obtaining the elevation image of the top of the target building: Using a computer to obtain building top elevation data from the generated building ‌DSM model, and using the building top elevation data as the building top elevation image; Step 3: Detection of target building elevation changes: Step 301: Obtain the building top elevation image before and after the damage according to the method of step 1 and step 2, and record them as the building top elevation image before the damage and the building top elevation image after the damage respectively; Step 302: using a computer to register the elevation image of the top of the building before damage and the elevation image of the top of the building after damage and perform difference calculation to obtain the elevation change area of ​​the target building; Step 4: Detection of damage on the top of the target building: Step 401: Obtain an image including the top of a target building taken from a bird's-eye view from a plurality of building images, and record the images including the top of the target building before and after damage as a building top image before damage and a building top image after damage, respectively. Step 402: using a computer to register the image of the top of the building before damage and the image of the top of the building after damage, and processing them through CVA algorithm and fuzzy C-means clustering algorithm to obtain the damaged area on the top of the target building; Step 5: Detection of damage to the side of the target building: Step 501: acquiring an image including the side of a target building taken from a side view from a plurality of building images, and recording the images including the side of the target building before and after damage as a building side image before damage and a building side image after damage, respectively; Step 502: Using a computer and using a SIFT feature matching method, register the side image of the building before damage and the side image of the building after damage to obtain a registered side image of the building before damage and a registered side image of the building after damage; Step 503: using a computer to perform empirical mode decomposition and reconstruction on the registered pre-damage building side image and the registered post-damage building side image respectively using a two-dimensional empirical mode decomposition method to obtain a reconstructed image of the pre-damage building side image with shadows removed and a reconstructed image of the post-damage building side image with shadows removed; Step 504: extract HOG features and perform Euclidean distance judgment on the reconstructed image of the building side before the damage and the reconstructed image of the building side after the damage to obtain the damaged area of ​​the target building side; Step 6: Visualization of target building damage: Step 601: Obtain a building image including the top and side of a target building from a plurality of building images and record it as the building image to be displayed; Step 602: Using a computer, mark the target building elevation change area, the target building top damage area and the target building side damage area on the building image to be displayed, so as to realize the overall visualization of the target building damage.

[0007] The above-mentioned three-dimensional building damage detection method is characterized in that: step 1, the specific process is as follows: Step 101: Use a visible light imager mounted on an unmanned aerial vehicle to collect all-round multi-view images of a target building, obtain multiple building images covering the entire target building, and transmit the collected multiple building images and POS data corresponding to the multiple building images to a computer; Step 102: Use a computer to import the multiple building images and the POS data corresponding to the multiple building images into Agisoft PhotoScan software, and use Agisoft PhotoScan software to process the multiple building images and the POS data corresponding to the multiple building images to generate a building DSM model.

[0008] The above-mentioned three-dimensional building damage detection method is characterized in that: step 302, the specific process is as follows: Step 3021: using a computer and using an image matching algorithm based on FFT phase correlation to register the elevation image of the top of the building before damage and the elevation image of the top of the building after damage, to obtain a registered elevation image of the top of the building before damage and a registered elevation image of the top of the building after damage; Step 3022: using a computer to perform a difference operation on the registered elevation image of the top of the building before damage and the registered elevation image of the top of the building after damage to obtain an elevation difference image; Step 3023: using a computer to perform binarization processing on the elevation difference image according to the set elevation threshold, to obtain an elevation difference binarization image; wherein the pixel values ​​greater than the elevation threshold are set to 255, and the pixel values ​​less than the elevation threshold are set to 0; Step 3024: Use a computer to record the white area in the elevation difference binary image as the building elevation change area.

[0009] The above-mentioned three-dimensional building damage detection method is characterized in that: step 402, the specific process is as follows: Step 4021: Using a computer and using a SIFT feature matching method, the image of the top of the building before damage and the image of the top of the building after damage are registered to obtain a registered image of the top of the building before damage and a registered image of the top of the building after damage; Step 4022: Using a computer and using a CVA algorithm, perform variable intensity processing on the R component, the G component, and the B component in the registered image of the top of the building before damage and the registered image of the top of the building after damage, to obtain a difference map of the top of the target building; Step 4023, using a computer to perform cluster segmentation on the target building top difference map using a fuzzy C-means clustering algorithm to obtain a target building top difference map after cluster segmentation; wherein, the number of clusters is 2, and the target building top difference map after cluster segmentation includes 2 types of cluster regions; Step 4024: Use a computer to obtain the pixel means of the two types of clustering areas, and the clustering area with a larger pixel mean is the clustering change area, and it is recorded as the damaged area on the top of the target building.

[0010] The above-mentioned three-dimensional building damage detection method is characterized in that: step 503, the specific process is as follows: Step 5031: Decompose the registered pre-damage building silhouette image using a two-dimensional empirical mode decomposition method using a computer to obtain N IMF components and a trend term; N is a positive integer; Step 5032: using a computer to obtain a grayscale histogram corresponding to the i-th IMF component according to the i-th IMF component; wherein i is a positive integer and 1≤i≤N; Step 5033: using a computer to obtain the ratio of the number of pixels less than the grayscale threshold to the total pixels from the grayscale histogram corresponding to the i-th IMF component; if the ratio of the number of pixels less than the grayscale threshold to the total pixels is greater than the ratio threshold, the shadow component of the IMF component is concentrated, and the IMF component is removed; Step 5034: Reconstruct the IMF components and a trend term after removing the shadow components using a computer to obtain a reconstructed image of the side of the building before damage; Step 5035: using a computer to obtain a shadow area from the registered side image of the building before damage, and obtaining a reconstructed image corresponding to the shadow area from the reconstructed image of the side of the building before damage, recorded as a regional reconstructed image; Step 5036: Using a computer, replace the shadow area with the regional reconstructed image on the registered side image of the building before the damage, thereby obtaining a reconstructed side image of the building before the damage with the shadow removed; Step 5037: According to the method of steps 5031 to 5036, perform empirical mode decomposition and reconstruction on the registered damaged building side image to obtain a reconstructed image of the damaged building side with shadows eliminated.

[0011] The above-mentioned three-dimensional building damage detection method is characterized in that: step 504, the specific process is as follows: Step 5041: Using a computer, with a window size of 3×3 and a sliding step size of 1, perform HOG feature extraction on the reconstructed image of the side of the building before the damage with the shadow removed, and obtain the HOG feature vector descriptor of the central pixels of M windows before the damage; wherein M is a positive integer and represents the total number of windows during HOG feature extraction; Step 5042: normalize the HOG feature vector descriptor of the j-th window center pixel before damage by using a computer to obtain the HOG normalized feature vector of the j-th window center pixel before damage; wherein j is a positive integer and 1≤j≤M; Step 5043: Use a computer to extract and normalize the HOG feature of the reconstructed image of the damaged building side with the shadow removed, and obtain the HOG normalized feature vector of the j-th window center pixel after the damage; Step 5044: Obtain the Euclidean distance dj between the HOG normalized feature vector of the j-th window center pixel before damage and the HOG normalized feature vector of the j-th window center pixel after damage; Step 5045: Use a computer to judge dj and the Euclidean distance threshold. If dj is greater than the Euclidean distance threshold, the central pixel of the j-th window is a changed pixel. Step 5046: Use a computer to take the connected area formed by the changed pixels as the damaged area on the side of the target building.

[0012] The above-mentioned three-dimensional building damage detection method is characterized in that: in step 5035, a computer is used to obtain the shadow area from the registered pre-damage building side image, as follows: Step A: Use a computer to transform the RGB color space corresponding to the registered pre-damage building side image into the HSI color space to obtain the corresponding hue component , Saturation component and brightness component ; Step B: Use a computer to calculate the , and get the ratio image ; Use the maximum inter-class variance method to compare the value image Processing to obtain a single threshold ; and obtain the ratio image The pixel value is greater than the single threshold Pixels of Step C: Using a computer to identify the image of the building before damage that is greater than a single threshold The area corresponding to the pixel point is used as the shadow segmentation image ; Step D: Obtaining shadow segmentation image using a computer The corresponding B channel component image is processed using the maximum inter-class variance method to obtain a single threshold ; and obtain the pixel value in the B channel component image that is less than a single threshold Pixels of Step E: Use a computer to segment the image in the shadow The medium will be less than the single threshold The area corresponding to the pixel point is used as the shadow segmentation image ; Step F: Obtaining shadow segmentation image using a computer The corresponding saturation component, and obtain the saturation value component greater than the set saturation threshold Pixels of Step G: Use a computer to segment the image in the shadow The medium will be greater than the threshold The area corresponding to the pixel point is processed as a connected area as the shadow area.

[0013] Compared with the prior art, the present invention has the following advantages: 1. The method of the present invention has simple steps and reasonable design, and solves the problem that the current building damage detection method is difficult to obtain comprehensive three-dimensional damage information of buildings.

[0014] 2. The present invention aligns the image of the top of the building before damage and the image of the top of the building after damage, and processes them through the CVA algorithm and the fuzzy C-means clustering algorithm to obtain the damaged area on the top of the target building. The CVA algorithm can simply and quickly perform change detection on the aligned two-phase images, thereby improving the detection speed.

[0015] 3. The present invention uses a two-dimensional empirical mode decomposition method to perform empirical mode decomposition and reconstruction on the registered building side images before and after damage, and performs HOG feature extraction and Euclidean distance judgment on this basis. Through empirical mode decomposition and reconstruction, the shadows of the target building before and after damage caused by different imaging conditions are eliminated as much as possible, thereby avoiding the influence of shadows on damage detection. By performing HOG feature extraction and normalization on the reconstructed image with shadow interference eliminated, the changes in the shape and structure of the target building before and after damage can be more accurately characterized.

[0016] 4. The present invention utilizes the elevation image of the top of the target building to detect the elevation change of the target building, and combines it with the detection of the damaged area on the top of the target building to adapt to the different degrees of damage on the top of the target building, and subsequently realizes the overall visualization of the damage to the target building, which is convenient for intuitive viewing.

[0017] In summary, the method of the present invention has simple steps and reasonable design. It uses an unmanned aerial vehicle to capture images of a target building from multiple angles, thereby detecting elevation changes, top damage and side damage of the target building, and realizing overall visualization of the damage to the target building.

[0018] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The figure is a flowchart of the method of the present invention.

[0020] Figure 2 It is an overall visualization diagram of the damage to the target building of the present invention. DETAILED DESCRIPTION

[0021] like Figure 1 As shown, the three-dimensional building damage detection method of the present invention comprises the following steps: Step 1: Use a drone equipped with a visible light imager to collect images of the target building, obtain multiple building images covering the entire target building, and generate a building DSM model; Step 2: Obtaining the elevation image of the top of the target building: Using a computer to obtain building top elevation data from the generated building ‌DSM model, and using the building top elevation data as the building top elevation image; Step 3: Detection of target building elevation changes: Step 301: Obtain the building top elevation image before and after the damage according to the method of step 1 and step 2, and record them as the building top elevation image before the damage and the building top elevation image after the damage respectively; Step 302: using a computer to register the elevation image of the top of the building before damage and the elevation image of the top of the building after damage and perform difference calculation to obtain the elevation change area of ​​the target building; Step 4: Detection of damage on the top of the target building: Step 401: Obtain an image including the top of a target building taken from a bird's-eye view from a plurality of building images, and record the images including the top of the target building before and after damage as a building top image before damage and a building top image after damage, respectively. Step 402: using a computer to register the image of the top of the building before damage and the image of the top of the building after damage, and processing them through CVA algorithm and fuzzy C-means clustering algorithm to obtain the damaged area on the top of the target building; Step 5: Detection of damage to the side of the target building: Step 501: acquiring an image including the side of a target building taken from a side view from a plurality of building images, and recording the images including the side of the target building before and after damage as a building side image before damage and a building side image after damage, respectively; Step 502: Using a computer and using a SIFT feature matching method, register the side image of the building before damage and the side image of the building after damage to obtain a registered side image of the building before damage and a registered side image of the building after damage; Step 503: using a computer to perform empirical mode decomposition and reconstruction on the registered pre-damage building side image and the registered post-damage building side image respectively using a two-dimensional empirical mode decomposition method to obtain a reconstructed image of the pre-damage building side image with shadows removed and a reconstructed image of the post-damage building side image with shadows removed; Step 504: extract HOG features and perform Euclidean distance judgment on the reconstructed image of the building side before the damage and the reconstructed image of the building side after the damage to obtain the damaged area of ​​the target building side; Step 6: Visualization of target building damage: Step 601: Obtain a building image including the top and side of a target building from a plurality of building images and record it as the building image to be displayed; Step 602: Using a computer, mark the target building elevation change area, the target building top damage area and the target building side damage area on the building image to be displayed, so as to realize the overall visualization of the target building damage.

[0022] In this embodiment, step one, the specific process is as follows: Step 101: Use a visible light imager mounted on an unmanned aerial vehicle to collect all-round multi-view images of a target building, obtain multiple building images covering the entire target building, and transmit the collected multiple building images and POS data corresponding to the multiple building images to a computer; Step 102: Use a computer to import the multiple building images and the POS data corresponding to the multiple building images into Agisoft PhotoScan software, and use Agisoft PhotoScan software to process the multiple building images and the POS data corresponding to the multiple building images to generate a building DSM model.

[0023] In this embodiment, step 302, the specific process is as follows: Step 3021: using a computer and using an image matching algorithm based on FFT phase correlation to register the elevation image of the top of the building before damage and the elevation image of the top of the building after damage, to obtain a registered elevation image of the top of the building before damage and a registered elevation image of the top of the building after damage; Step 3022: using a computer to perform a difference operation on the registered elevation image of the top of the building before damage and the registered elevation image of the top of the building after damage to obtain an elevation difference image; Step 3023: using a computer to perform binarization processing on the elevation difference image according to the set elevation threshold, to obtain an elevation difference binarization image; wherein the pixel values ​​greater than the elevation threshold are set to 255, and the pixel values ​​less than the elevation threshold are set to 0; Step 3024: Use a computer to record the white area in the elevation difference binary image as the building elevation change area.

[0024] In this embodiment, the specific process of step 402 is as follows: Step 4021: Using a computer and using a SIFT feature matching method, the image of the top of the building before damage and the image of the top of the building after damage are registered to obtain a registered image of the top of the building before damage and a registered image of the top of the building after damage; Step 4022: Using a computer and using a CVA algorithm, perform variable intensity processing on the R component, the G component, and the B component in the registered image of the top of the building before damage and the registered image of the top of the building after damage, to obtain a difference map of the top of the target building; Step 4023, using a computer to perform cluster segmentation on the target building top difference map using a fuzzy C-means clustering algorithm to obtain a target building top difference map after cluster segmentation; wherein, the number of clusters is 2, and the target building top difference map after cluster segmentation includes 2 types of cluster regions; Step 4024: Use a computer to obtain the pixel means of the two types of clustering areas, and the clustering area with a larger pixel mean is the clustering change area, and it is recorded as the damaged area on the top of the target building.

[0025] In this embodiment, step 503, the specific process is as follows: Step 5031: Decompose the registered pre-damage building silhouette image using a two-dimensional empirical mode decomposition method using a computer to obtain N IMF components and a trend term; N is a positive integer; Step 5032: using a computer to obtain a grayscale histogram corresponding to the i-th IMF component according to the i-th IMF component; wherein i is a positive integer and 1≤i≤N; Step 5033: using a computer to obtain the ratio of the number of pixels less than the grayscale threshold to the total pixels from the grayscale histogram corresponding to the i-th IMF component; if the ratio of the number of pixels less than the grayscale threshold to the total pixels is greater than the ratio threshold, the shadow component of the IMF component is concentrated, and the IMF component is removed; Step 5034: Reconstruct the IMF components and a trend term after removing the shadow components using a computer to obtain a reconstructed image of the side of the building before damage; Step 5035: using a computer to obtain a shadow area from the registered side image of the building before damage, and obtaining a reconstructed image corresponding to the shadow area from the reconstructed image of the side of the building before damage, recorded as a regional reconstructed image; Step 5036: Using a computer, replace the shadow area with the regional reconstructed image on the registered side image of the building before the damage, thereby obtaining a reconstructed side image of the building before the damage with the shadow removed; Step 5037: Perform empirical mode decomposition and reconstruction on the registered damaged building side image according to the method of steps 5031 to 5036 to obtain a reconstructed image of the damaged building side with shadows removed.

[0026] In this embodiment, step 504, the specific process is as follows: Step 5041: Using a computer, with a window size of 3×3 and a sliding step size of 1, perform HOG feature extraction on the reconstructed image of the side of the building before the damage with the shadow removed, and obtain the HOG feature vector descriptor of the central pixels of M windows before the damage; wherein M is a positive integer and represents the total number of windows during HOG feature extraction; Step 5042: normalize the HOG feature vector descriptor of the j-th window center pixel before damage by using a computer to obtain the HOG normalized feature vector of the j-th window center pixel before damage; wherein j is a positive integer and 1≤j≤M; Step 5043: Use a computer to extract and normalize the HOG feature of the reconstructed image of the damaged building side with the shadow removed, and obtain the HOG normalized feature vector of the j-th window center pixel after the damage; Step 5044: Obtain the Euclidean distance dj between the HOG normalized feature vector of the j-th window center pixel before damage and the HOG normalized feature vector of the j-th window center pixel after damage; Step 5045: Use a computer to judge dj and the Euclidean distance threshold. If dj is greater than the Euclidean distance threshold, the central pixel of the j-th window is a changed pixel. Step 5046: Use a computer to take the connected area formed by the changed pixels as the damaged area on the side of the target building.

[0027] In this embodiment, in step 5035, a computer is used to obtain the shadow area from the registered pre-damage building side image, as follows: Step A: Use a computer to transform the RGB color space corresponding to the registered pre-damage building side image into the HSI color space to obtain the corresponding hue component , Saturation component and brightness component ; Step B: Use a computer to calculate the , and get the ratio image ; Use the maximum inter-class variance method to compare the value image Processing to obtain a single threshold ; and obtain the ratio image The pixel value is greater than the single threshold Pixels of Step C: Using a computer to identify the image of the building before damage that is greater than a single threshold The area corresponding to the pixel point is used as the shadow segmentation image ; Step D: Obtaining shadow segmentation image using a computer The corresponding B channel component image is processed using the maximum inter-class variance method to obtain a single threshold ; and obtain the pixel value in the B channel component image that is less than a single threshold Pixels of Step E: Use a computer to segment the image in the shadow The medium will be less than the single threshold The area corresponding to the pixel point is used as the shadow segmentation image ; Step F: Obtaining shadow segmentation image using a computer The corresponding saturation component, and obtain the saturation value component greater than the set saturation threshold Pixels of Step G: Use a computer to segment the image in the shadow The medium will be greater than the threshold The area corresponding to the pixel point is processed as a connected area as the shadow area.

[0028] In this embodiment, during specific implementation, the heading overlap and the lateral overlap during image acquisition in step 101 are not less than 30%.

[0029] In this embodiment, during specific implementation, the elevation threshold in step 3023 is 3 meters, which can be adjusted according to actual needs.

[0030] In this embodiment, when implemented specifically, step 4022 is specifically: perform difference operations on the R component, G component and B component in the registered image of the top of the building before damage and the registered image of the top of the building after damage, respectively, to obtain the R component change value, G component change value and B component change value, and use the R component change value, G component change value and B component change value as a three-component vector, and then calculate the amplitude of the three-component vector as the change intensity, then the change intensity at each pixel point is the difference map of the top of the target building.

[0031] In this embodiment, when implementing, the saturation threshold is set The value is 30, which can be adjusted according to actual requirements In this embodiment, during specific implementation, the value of N in step 5031 is 5, and can be adjusted as needed during actual use.

[0032] In this embodiment, during specific implementation, the grayscale threshold in step 5033 is 20, and the ratio threshold is 10%. In actual use, they can be adjusted as needed.

[0033] In this embodiment, during specific implementation, the HOG feature is a gradient directional histogram feature.

[0034] In this embodiment, during specific implementation, the number of pixels in the connected area formed in step 5046 can be determined, and the connected area greater than the preset value is used as the damaged area to avoid the influence of isolated pixels. The preset value can be 50 and can be adjusted as needed.

[0035] In this embodiment, during specific implementation, when performing HOG feature extraction in step 5041, the size of each window is 3×3, and the HOG feature vector descriptor of the central pixel of each window is a one-dimensional feature vector of 8×1.

[0036] In this embodiment, during specific implementation, the Euclidean distance threshold in step 5045 is 0.1, and can be adjusted as needed during actual use.

[0037] In this embodiment, the CVA algorithm, namely the change vector detection method, can simply and quickly perform change detection on the registered image of the top of the building before damage and the registered image of the top of the building after damage.

[0038] In this embodiment, the method of the present invention is used to detect the elevation change, top damage and side damage of the target building, and the following is obtained: Figure 2 An overall visualization of the target building damage is shown.

[0039] In summary, the method of the present invention has simple steps and reasonable design. It uses an unmanned aerial vehicle to capture images of a target building from multiple angles, thereby detecting elevation changes, top damage and side damage of the target building, and realizing overall visualization of the damage to the target building.

[0040] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any simple modification, change and equivalent structural change made to the above embodiment based on the technical essence of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A three-dimensional building damage detection method, characterized in that: The method comprises the following steps: Step 1: Use a drone equipped with a visible light imager to collect images of the target building, obtain multiple building images covering the entire target building, and generate a building DSM model; Step 2: Obtaining the elevation image of the top of the target building: Using a computer to obtain building top elevation data from the generated building ‌DSM model, and using the building top elevation data as the building top elevation image; Step 3: Detection of target building elevation changes: Step 301: Obtain the building top elevation image before and after the damage according to the method of step 1 and step 2, and record them as the building top elevation image before the damage and the building top elevation image after the damage respectively; Step 302: using a computer to register the elevation image of the top of the building before damage and the elevation image of the top of the building after damage and perform difference calculation to obtain the elevation change area of ​​the target building; Step 4: Detection of damage on the top of the target building: Step 401: Obtain an image including the top of a target building taken from a bird's-eye view from a plurality of building images, and record the images including the top of the target building before and after damage as a building top image before damage and a building top image after damage, respectively. Step 402: using a computer to register the image of the top of the building before damage and the image of the top of the building after damage, and processing them through CVA algorithm and fuzzy C-means clustering algorithm to obtain the damaged area on the top of the target building; Step 5: Detection of damage to the side of the target building: Step 501: acquiring an image including the side of a target building taken from a side view from a plurality of building images, and recording the images including the side of the target building before and after damage as a building side image before damage and a building side image after damage, respectively; Step 502: Using a computer and using a SIFT feature matching method, register the side image of the building before damage and the side image of the building after damage to obtain a registered side image of the building before damage and a registered side image of the building after damage; Step 503: using a computer to perform empirical mode decomposition and reconstruction on the registered pre-damage building side image and the registered post-damage building side image respectively using a two-dimensional empirical mode decomposition method to obtain a reconstructed image of the pre-damage building side image with shadows removed and a reconstructed image of the post-damage building side image with shadows removed; Step 504: extract HOG features and perform Euclidean distance judgment on the reconstructed image of the building side before the damage and the reconstructed image of the building side after the damage to obtain the damaged area of ​​the target building side; Step 6: Visualization of target building damage: Step 601: Obtain a building image including the top and side of a target building from a plurality of building images and record it as the building image to be displayed; Step 602: Using a computer, mark the target building elevation change area, the target building top damage area and the target building side damage area on the building image to be displayed, so as to realize the overall visualization of the target building damage.

2. A three-dimensional building damage detection method according to claim 1, characterized in that: Step 1: The specific process is as follows: Step 101: Use a visible light imager mounted on an unmanned aerial vehicle to collect all-round multi-view images of a target building, obtain multiple building images covering the entire target building, and transmit the collected multiple building images and POS data corresponding to the multiple building images to a computer; Step 102: Use a computer to import the multiple building images and the POS data corresponding to the multiple building images into Agisoft PhotoScan software, and use Agisoft PhotoScan software to process the multiple building images and the POS data corresponding to the multiple building images to generate a building DSM model.

3. A three-dimensional building damage detection method according to claim 1, characterized in that: Step 302, the specific process is as follows: Step 3021: using a computer and using an image matching algorithm based on FFT phase correlation to register the elevation image of the top of the building before damage and the elevation image of the top of the building after damage, to obtain a registered elevation image of the top of the building before damage and a registered elevation image of the top of the building after damage; Step 3022: using a computer to perform a difference operation on the registered elevation image of the top of the building before damage and the registered elevation image of the top of the building after damage to obtain an elevation difference image; Step 3023: using a computer to perform binarization processing on the elevation difference image according to the set elevation threshold, to obtain an elevation difference binarization image; wherein the pixel values ​​greater than the elevation threshold are set to 255, and the pixel values ​​less than the elevation threshold are set to 0; Step 3024: Use a computer to record the white area in the elevation difference binary image as the building elevation change area.

4. A three-dimensional building damage detection method according to claim 1, characterized in that: Step 402: The specific process is as follows: Step 4021: Using a computer and using a SIFT feature matching method, the image of the top of the building before damage and the image of the top of the building after damage are registered to obtain a registered image of the top of the building before damage and a registered image of the top of the building after damage; Step 4022: Using a computer and using a CVA algorithm, perform variable intensity processing on the R component, the G component, and the B component in the registered image of the top of the building before damage and the registered image of the top of the building after damage, to obtain a difference map of the top of the target building; Step 4023, using a computer to perform cluster segmentation on the target building top difference map using a fuzzy C-means clustering algorithm to obtain a target building top difference map after cluster segmentation; wherein, the number of clusters is 2, and the target building top difference map after cluster segmentation includes 2 types of cluster regions; Step 4024: Use a computer to obtain the pixel means of the two types of clustering areas, and the clustering area with a larger pixel mean is the clustering change area, and it is recorded as the damaged area on the top of the target building.

5. A three-dimensional building damage detection method according to claim 1, characterized in that: Step 503, the specific process is as follows: Step 5031: Decompose the registered pre-damage building silhouette image using a two-dimensional empirical mode decomposition method using a computer to obtain N IMF components and a trend term; N is a positive integer; Step 5032: using a computer to obtain a grayscale histogram corresponding to the i-th IMF component according to the i-th IMF component; wherein i is a positive integer and 1≤i≤N; Step 5033: using a computer to obtain the ratio of the number of pixels less than the grayscale threshold to the total pixels from the grayscale histogram corresponding to the i-th IMF component; if the ratio of the number of pixels less than the grayscale threshold to the total pixels is greater than the ratio threshold, the shadow component of the IMF component is concentrated, and the IMF component is removed; Step 5034: Reconstruct the IMF components and a trend term after removing the shadow components using a computer to obtain a reconstructed image of the side of the building before damage; Step 5035: using a computer to obtain a shadow area from the registered side image of the building before damage, and obtaining a reconstructed image corresponding to the shadow area from the reconstructed image of the side of the building before damage, recorded as a regional reconstructed image; Step 5036: Using a computer, replace the shadow area with the regional reconstructed image on the registered side image of the building before the damage, thereby obtaining a reconstructed side image of the building before the damage with the shadow removed; Step 5037: According to the method of steps 5031 to 5036, perform empirical mode decomposition and reconstruction on the registered damaged building side image to obtain a reconstructed image of the damaged building side with shadows eliminated.

6. A three-dimensional building damage detection method according to claim 1, characterized in that: Step 504, the specific process is as follows: Step 5041: Using a computer, with a window size of 3×3 and a sliding step size of 1, perform HOG feature extraction on the reconstructed image of the side of the building before the damage with the shadow removed, and obtain the HOG feature vector descriptor of the central pixels of M windows before the damage; wherein M is a positive integer and represents the total number of windows during HOG feature extraction; Step 5042: normalize the HOG feature vector descriptor of the j-th window center pixel before damage by using a computer to obtain the HOG normalized feature vector of the j-th window center pixel before damage; wherein j is a positive integer and 1≤j≤M; Step 5043: Use a computer to extract and normalize the HOG feature of the reconstructed image of the damaged building side with the shadow removed, and obtain the HOG normalized feature vector of the j-th window center pixel after the damage; Step 5044: Obtain the Euclidean distance dj between the HOG normalized feature vector of the j-th window center pixel before damage and the HOG normalized feature vector of the j-th window center pixel after damage; Step 5045: Use a computer to judge dj and the Euclidean distance threshold. If dj is greater than the Euclidean distance threshold, the central pixel of the j-th window is a changed pixel. Step 5046: Use a computer to take the connected area formed by the changed pixels as the damaged area on the side of the target building.

7. A three-dimensional building damage detection method according to claim 1, characterized in that: In step 5035, a computer is used to obtain the shadow area from the registered pre-damage building silhouette image, as follows: Step A: Use a computer to transform the RGB color space corresponding to the registered pre-damage building side image into the HSI color space to obtain the corresponding hue component , Saturation component and brightness component ; Step B: Use a computer to calculate the , and get the ratio image ; Use the maximum inter-class variance method to compare the value image Processing to obtain a single threshold ; And get the ratio image The pixel value is greater than the single threshold Pixels of Step C: Using a computer to identify the image of the building before damage that is greater than a single threshold The area corresponding to the pixel point is used as the shadow segmentation image ; Step D: Obtaining shadow segmentation image using a computer The corresponding B channel component image is processed using the maximum inter-class variance method to obtain a single threshold ; And get the pixel value in the B channel component image that is less than a single threshold Pixels of Step E: Use a computer to segment the image in the shadow The medium will be less than the single threshold The area corresponding to the pixel point is used as the shadow segmentation image ; Step F: Obtaining shadow segmentation image using a computer The corresponding saturation component, and obtain the saturation value component greater than the set saturation threshold Pixels of Step G: Use a computer to segment the image in the shadow The medium will be greater than the threshold The area corresponding to the pixel point is processed as a connected area as the shadow area.

Images