A method for extracting urban green areas based on drones

By acquiring images from multiple drones at multiple altitudes, performing distortion correction, and stitching images together, the accuracy problem of extracting green areas from drone aerial photography has been solved, achieving efficient and precise extraction of green areas.

CN118411380BActive Publication Date: 2024-11-22ZHEJIANG DIXIN SOFTWARE TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410593937.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-14
Publication Date
2024-11-22
Estimated Expiration
2044-05-14

AI Technical Summary

Technical Problem

Existing methods for extracting urban green spaces using drone aerial photography suffer from image distortion and insufficient accuracy, especially under conditions of transmission delay or signal interference, which affects the accuracy of green space extraction.

Method used

Multiple drones are used to capture images at different altitudes, and distortion and geometric corrections are performed. Combined with image stitching and contour recognition, errors are automatically eliminated through image stacking algorithms and model learning, so as to achieve accurate extraction of green areas.

Benefits of technology

It improves the accuracy and efficiency of green area extraction, reduces image distortion and human intervention, and enhances the stability and precision of the system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118411380B_ABST
    Figure CN118411380B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for extracting urban green areas based on drones, comprising the following steps: step one, selecting an area where green areas need to be extracted, dividing the area into measurement areas, selecting a point in each divided measurement area and placing drones one by one; step two, starting each drone, and each drone takes off to a shooting height; step three, performing distortion processing on the acquired original image; and step four, splicing each image based on the image features at the edge of the image identified in step three. The method for extracting urban green areas based on drones of the present invention can extract urban green areas by taking off and shooting multiple drones through the settings of steps one to four, and then splicing the images. Compared with the method in the prior art, there is no continuous image transmitted in real time by drones, so it is not easy to have image distortion.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a method for extracting urban green areas, and more particularly to a method for extracting urban green areas based on unmanned aerial vehicles. Background Art

[0002] Urban green space refers to the green space within the urban construction land and the square land and the regional green space outside the urban construction land. It contains two levels of content: one is the land used for greening within the scope of urban construction land; the other is outside the urban construction land. Urban greening can improve the quality of urban natural ecology and is beneficial to environmental protection; improve the quality of urban life and adjust environmental psychology; increase the aesthetic effect of urban landscape; increase urban economic benefits; be beneficial to urban disaster prevention; and purify air pollution.

[0003] Traditional urban green space measurement methods mainly rely on a lot of manpower and time for on-site investigation and survey. Although traditional measurement methods are highly accurate, they require a lot of manpower and time. In the existing method of extracting urban green spaces through drone aerial photography, due to the limitations of aerial photography height and technology, the orthophoto effect is not ideal, and later extraction still requires a lot of manual intervention.

[0004] To this end, there is currently an invention patent in the prior art with patent number 202111371224.X, entitled "Urban Green Space Extraction Method Based on UAV Aerial Images", which discloses obtaining original images through a drone, and then processing the obtained original images to finally achieve the extraction of urban green spaces. However, in the above-mentioned green space extraction method, the images taken by the drone are shot at a fixed height, so the area that the collected original images can cover is not large. Therefore, the drone is required to fly around the extraction area, and then the green area is extracted by recognizing continuous images. However, since the continuous images need to be transmitted by the drone in real time, if there is a transmission delay or the signal is interfered with, image distortion is likely to occur, which seriously affects the accuracy of the extracted urban green space. Summary of the invention

[0005] In view of the shortcomings of the prior art, the object of the present invention is to provide a method for extracting urban green areas based on drones, which can effectively ensure the accuracy of the extracted urban green spaces.

[0006] To achieve the above object, the present invention provides the following technical solution: a method for extracting urban green areas based on drones, characterized in that it comprises the following steps:

[0007] Step 1: Select the area where green area extraction is required, divide the area into measurement areas, and select a point in each divided measurement area to place the drone;

[0008] Step 2: Start each drone, take off to a shooting height, and shoot an image of the survey area after the drone reaches the shooting height to obtain an original image, wherein the range of the image shot by the drone is larger than the survey area when the drone reaches the shooting height;

[0009] Step 3: The original image is distorted and processed. The rectified image is processed into an aerial image with coordinate information. Then, geometric correction and uniform light and color processing are performed to finally obtain an orthophoto result map. Image recognition is performed on the obtained orthophoto result map to identify and extract the green area in the image and the image features at the edge of the image.

[0010] Step 4: Based on the image features at the edge of the image identified in step 3, each image is stitched together to form a large image. The green area combination identified and extracted by each orthophoto result map constitutes the extracted urban green area.

[0011] As a further improvement of the present invention, in the step 2, each drone takes a plurality of original images, and then all the original images obtained above are distorted through step 3. The corrected image is processed into an aerial image with coordinate information, and then geometric correction and uniform light and color processing are performed to finally obtain an orthophoto result map. Image recognition is performed on each obtained orthophoto result map to identify and extract the green area in the image, obtain the outline of the green area, and then compare the outline of the green area in each orthophoto result map with each other, eliminate the outline of the green area that has changed, and finally retain the outline of the green area that has not changed as the green area in the image finally identified and extracted.

[0012] As a further improvement of the present invention, in the step 2, the shooting heights of each drone taking off are set to be multiple from low to high, and in the step 3, after obtaining the orthophoto result map, an image refinement step is also performed, and the refinement step is as follows:

[0013] Step 31: drive the drone to descend to the lowest shooting height, then scale down the image taken at the lowest height and overlay it on the image taken at the second lowest height with the same center, and spread the image taken at the lowest height over the image taken at the second lowest height by copying and spreading it, and calculate and obtain the center point of the image taken at the lowest height;

[0014] Step 32: driving the drone to move to each center point obtained in step 31 in sequence at the lowest shooting altitude, and taking an image when moving to a center point;

[0015] Step 33, scale down the image taken by the drone at the center point, and then replace the original image corresponding to the center point until the newly taken image completely replaces the old image corresponding to the center point, so as to cover the image taken at the second lowest altitude, and then fuse the stacked images through the image stacking algorithm to complete the refinement of the second lowest altitude image. Before comparing the outlines of the green areas of each orthophoto result map in step 2, scale down the outlines of each green area from low to high according to the shooting height, and then overlap the outlines of each green area, and remove the outlines without two or more overlaps.

[0016] As a further improvement of the present invention, the outline of each green area is reduced in proportion from low to high according to the shooting height as follows: an area with a clearly distinguishable outline is set, and then the area with a clear outline is photographed by a drone according to various shooting heights, and each original image is collected. Thereafter, the outline is identified, and the size change of the identified outline is used as training data to build a model. The model is used to learn the reduction ratio of the outline as the shooting height increases. After the model training is completed, the image taken by the drone is directly input into the model for scale reduction.

[0017] As a further improvement of the present invention, the specific steps of stitching the images in step 4 are as follows:

[0018] Step 41: connect the drone placement points selected in step 1 diagonally, and construct a rectangular coordinate system with the intersection of the diagonal lines as the origin;

[0019] Step 42: Place each orthophoto result map into a rectangular coordinate system. The center point of each orthophoto result map coincides with the drone placement point. Perform contour alignment and fusion on the overlapping edges of each orthophoto result map to fuse and stitch the edges of each orthophoto result map into an overall large image.

[0020] As a further improvement of the present invention, the specific method of fusing images through the image stacking algorithm in the step three is: before the newly captured image is scaled down, the outline of the green area in the image is identified; after the outline of the green area and the image are scaled down together, the image is made transparent while the outline of the green area is maintained, and then superimposed on the image captured at the second lowest height, and overlapped with the outline of the green area identified in the image captured at the second lowest height, and the outline of the green area identified in the original image captured at the second lowest height is replaced.

[0021] As a further improvement of the present invention, in the process of stacking green area outlines, if there is a green area outline that cannot be stacked, the outline is retained and marked in red, and then the stacked image is sent to the human-computer interaction device for manual judgment and processing of the green area outline that cannot be stacked. As a further improvement of the present invention, the measurement area range divided in step 1 is a square, and the point where the drone is placed is selected within the measurement area as the center point of the square.

[0022] The beneficial effect of the present invention is that through the setting of step one, the survey area range can be effectively selected, and the shooting location of each drone can be selected at the same time. Then, through the setting of step two, the image of the entire area can be effectively captured by a combination of multiple drones. Then, through the setting of steps three and four, the green area in the image of the entire area can be simply and effectively extracted, thereby completing the extraction of the green area for the area. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A schematic diagram of a top view taken by a drone of the method for extracting urban green areas based on a drone of the present invention;

[0024] Figure 2 This is a schematic diagram of the main view taken by a drone of the method for extracting urban green areas based on a drone of the present invention. DETAILED DESCRIPTION

[0025] The present invention will be further described below in detail with reference to the embodiments shown in the accompanying drawings.

[0026] Reference Figures 1 to 2 As shown, a method for extracting urban green areas based on a drone in this embodiment is characterized by comprising the following steps:

[0027] Step 1: Select the area where green area extraction is required, divide the area into measurement areas, and select a point in each divided measurement area to place the drone;

[0028] Step 2: Start each drone, take off to a shooting height, and shoot an image of the survey area after the drone reaches the shooting height to obtain an original image, wherein the range of the image shot by the drone is larger than the survey area when the drone reaches the shooting height;

[0029] Step 3: The original image is distorted and processed. The rectified image is processed into an aerial image with coordinate information. Then, geometric correction and uniform light and color processing are performed to finally obtain an orthophoto result map. Image recognition is performed on the obtained orthophoto result map to identify and extract the green area in the image and the image features at the edge of the image.

[0030] Step 4: splice each image based on the image features at the edge of the image identified in step 3, and splice each orthophoto result map into a large image. The green area combination identified and extracted by each orthophoto result map constitutes the extracted urban green area. In the process of using the extraction method of this embodiment, it is only necessary to execute steps 1 to 4 in sequence. Compared with the method in the prior art, the area to be extracted can be divided into multiple measurement areas, and then each measurement area is taken by flying a drone to a take-off height and then shooting. The original image of each measurement area can be stably collected. After that, through the effects of steps 2, 3 and 4, the original image is processed, the green area in each image is extracted, and then the green area is spliced ​​into a whole image to realize the extraction of the green area of ​​the entire city. The measurement area in this embodiment is a square, which is well matched with the shape of the original image that can be taken by the drone to realize the extraction of the green area of ​​the city. Compared with the method of using only a single drone in the prior art, it does not require long-term transmission of continuous images, reducing the occurrence of image distortion.

[0031] As a specific implementation method of the improvement, in the step 2, each drone takes a plurality of original images, and then all the original images obtained above are distorted through step 3. The corrected image is processed into an aerial image with coordinate information, and then geometric correction and uniform light and color processing are performed to finally obtain an orthophoto result map. Image recognition is performed on each obtained orthophoto result map to identify and extract the green area in the image, obtain the outline of the green area, and then compare the outline of the green area of ​​each orthophoto result map with each other, eliminate the outline of the green area with changes, and finally retain the outline of the green area without changes as the green area in the final identified and extracted image. Greening area, since the outline of the greening area in the city basically does not change, the most commonly used image recognition algorithm is the RGB recognition algorithm, which classifies the various features on the image by identifying the color difference of the image. Therefore, in the process of identifying the greening area of ​​the original image, if the color of other objects (such as vehicles) is close to the color of the greening area, they are easily identified and classified into the greening area together, resulting in certain errors. Therefore, by taking multiple original images and then comparing the multiple original images, the changing outline can be effectively identified, which can effectively realize the automatic exclusion of the above-mentioned other objects and reduce the manual workload.

[0032] As a specific implementation of the improvement, in the step 2, the shooting heights of each drone taking off are set to be multiple from low to high, and in the step 3, after obtaining the orthophoto result map, an image refinement step is also performed, and the refinement step is as follows:

[0033] Step 31: drive the drone to descend to the lowest shooting height, then scale down the image taken at the lowest height and overlay it on the image taken at the second lowest height with the same center, and spread the image taken at the lowest height over the image taken at the second lowest height by copying and spreading it, and calculate and obtain the center point of the image taken at the lowest height;

[0034] Step 32: driving the drone to move to each center point obtained in step 31 in sequence at the lowest shooting altitude, and taking an image when moving to a center point;

[0035] Step 33, scale down the image taken by the drone at the center point, and then replace the original image corresponding to the center point until the newly taken image completely replaces the old image corresponding to the center point, so as to cover the image taken at the second lowest altitude, and then fuse the stacked images through the image stacking algorithm to complete the refinement of the second lowest altitude image. Before comparing the outlines of the green areas of each orthophoto result map in step 2, scale down the outlines of each green area from low to high according to the shooting height, and then overlap the outlines of each green area, and remove the outlines without two or more overlaps. Through the settings of the above steps 31 to 33, it is possible to effectively use the image taken at a higher shooting height as the base image, and then take multiple images at a lower shooting height as supplementary images. After that, the base image is supplemented by splicing the supplementary images. In this way, the image taken at a lower height has a clearer green area outline to replace the green area outline on the base image. In this way, the green area outline of the image with a higher shooting height is refined in a simple and effective way, so that the green area outline finally obtained is larger in area and has a higher outline accuracy.

[0036] As a specific implementation method of the improvement, the outline of each green area is reduced in proportion from low to high according to the shooting height as follows: an area with a clearly distinguishable outline is set, and then the area with a clear outline is photographed by a drone at various shooting heights, and each original image is collected. The outline is then identified, and the changes in the size of the identified outline are used as training data to build a model. The model is used to learn the reduction ratio of the outline as the shooting height increases. After the model training is completed, the image taken by the drone is directly input into the model for scale reduction. In the above manner, the image can be scaled down through the AI ​​learning model without the need to manually set a large number of parameters, which effectively reduces the manual workload.

[0037] As a specific implementation method of the improvement, the specific steps of stitching the images in step 4 are as follows: step 41, diagonally connecting the drone placement points selected in step 1, and constructing a rectangular coordinate system with the intersection of the diagonal lines as the origin;

[0038] Step 42: Place each orthophoto result map into a rectangular coordinate system. The center point of each orthophoto result map coincides with the drone placement point. Perform contour alignment and fusion on the overlapping edges of each orthophoto result map to fuse and splice the edges of each orthophoto result map to form an overall large image. By setting the above steps, image stitching can be achieved simply and effectively. In the process of stitching, the image edges are stitched together through the characteristic contours, so the stitched image has better completeness.

[0039] As an improved specific implementation method, the specific method of fusing images through the image stacking algorithm in the step three is: before the newly captured image is scaled down, the outline of the green area in the image is identified; after the outline of the green area and the image are scaled down together, the image is made transparent while the outline of the green area is maintained, and then superimposed on the image captured at the second lowest height, and overlapped with the outline of the green area identified in the image captured at the second lowest height, and the outline of the green area identified in the original image captured at the second lowest height is replaced. Through the setting of the above steps, the stacking algorithm can be effectively used to fuse the images, and the overall implementation process is simple and convenient.

[0040] As a specific implementation method of the improvement, in the process of stacking green area outlines, if there is a green area outline that cannot be stacked, the outline will be retained and marked in red. The stacked image will then be sent to a human-computer interaction device for manual judgment of the green area outline that cannot be stacked. Through the setting of the above steps, further recognition and judgment of the green area outline can be effectively achieved, and the use of manual assisted judgment can effectively improve the accuracy of the judgment.

[0041] To sum up, the urban green area extraction method based on drones in this embodiment, through the settings of steps one to four, can effectively realize the use of multiple drones to synchronously collect original images of multiple survey areas, and then realize the extraction of large-scale green areas by splicing the original images. Compared with the method of continuously shooting and collecting images by mobile drones in the prior art, there is no need to ensure the continuity of data, thereby reducing the impact of signal transmission delay and interference.

[0042] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A method for extracting urban green areas based on drones, characterized in that: The steps include: Step 1: Select the area where green area extraction is required, divide the area into measurement areas, and select a point in each divided measurement area to place the drone; Step 2: Start each drone, take off to a shooting height, and shoot an image of the survey area after the drone reaches the shooting height to obtain an original image, wherein the range of the image shot by the drone is larger than the survey area when the drone reaches the shooting height; Step 3: The original image is distorted and processed. The rectified image is processed into an aerial image with coordinate information. Then, geometric correction and uniform light and color processing are performed to finally obtain an orthophoto result map. Image recognition is performed on the obtained orthophoto result map to identify and extract the green area in the image and the image features at the edge of the image. Step 4: splice each image based on the image features at the edge of the image identified in step 3, and splice each orthophoto result map into a large image. The green area combination identified and extracted by each orthophoto result map constitutes the extracted urban green area; in the step 2, each drone takes a plurality of original images, and then all the original images obtained above are distorted through step 3. The corrected image will be processed into an aerial image map with coordinate information, and then geometric correction and uniform light and color processing are performed to finally obtain the orthophoto result map, and image recognition is performed on each obtained orthophoto result map to identify and extract the green area in the image, obtain the outline of the green area, and then compare the outline of the green area of ​​each orthophoto result map with each other, eliminate the outline of the green area with changes, and finally retain the outline of the green area without changes as the green area in the image finally identified and extracted; in the step 2, the shooting heights of each drone taking off are set from low to high, and in the step 3, after obtaining the orthophoto result map, an image refinement step is also performed, and the refinement step is as follows: Step 31: drive the drone to descend to the lowest shooting height, then scale down the image taken at the lowest height and overlay it on the image taken at the second lowest height with the same center, and spread the image taken at the lowest height over the image taken at the second lowest height by copying and spreading it, and calculate and obtain the center point of the image taken at the lowest height; Step 32: driving the drone to move to each center point obtained in step 31 in sequence at the lowest shooting altitude, and taking an image when moving to a center point; Step 33, scale down the image taken by the drone at the center point, and then replace the original image corresponding to the center point until the newly taken image completely replaces the old image corresponding to the center point to cover the image taken at the second lowest height, and then fuse the stacked images through the image stacking algorithm to complete the refinement of the second lowest height image. Before comparing the outlines of the green areas of each orthophoto result map with each other in step 2, scale down the outlines of each green area obtained from low to high according to the shooting height, and then overlap the outlines of each green area, and remove the outlines without two or more overlaps; the outlines of each green area are scaled down from low to high according to the shooting height as follows: set an area with obvious contour distinction, and then use the drone to shoot the area with obvious contour according to each shooting height, collect each original image, and then identify the contour, use the identified contour size change as training data to build a model, learn the reduction ratio of the contour as the shooting height increases through the model, and after the model training and learning is completed, directly input the image taken by the drone into the model for scale reduction.

2. The method for extracting urban green areas based on drones according to claim 1 is characterized in that: The specific steps of stitching the images in step 4 are as follows: Step 41: connect the drone placement points selected in step 1 diagonally, and construct a rectangular coordinate system with the intersection of the diagonal lines as the origin; Step 42: Place each orthophoto result map into a rectangular coordinate system. The center point of each orthophoto result map coincides with the drone placement point. Perform contour alignment and fusion on the overlapping edges of each orthophoto result map to fuse and stitch the edges of each orthophoto result map into an overall large image.

3. The method for extracting urban green areas based on drones according to claim 2 is characterized in that: The specific method of fusing images through the image stacking algorithm in the step three is: before the newly captured image is scaled down, the outline of the green area in the image is identified; after the outline of the green area and the image are scaled down together, the image is made transparent while the outline of the green area is maintained, and then superimposed on the image captured at the second lowest height, and overlapped with the outline of the green area identified in the image captured at the second lowest height, and the outline of the green area identified in the original image captured at the second lowest height is replaced.

4. The method for extracting urban green areas based on drones according to claim 3 is characterized in that: In the process of stacking green area outlines, if there is a green area outline that cannot be stacked, the outline will be retained and marked in red. The stacked image will then be sent to a human-computer interaction device for manual judgment and processing of the green area outline that cannot be stacked.

5. The method for extracting urban green areas based on drones according to any one of claims 1 to 4, characterized in that: The measurement areas divided in step 1 are all squares, and the point where the drone is placed is selected within the measurement area as the center point of the square.