Wildlife identification and positioning method and system based on oblique photogrammetry

Abstract

This invention relates to the field of aerial photogrammetry technology and proposes a method and system for wildlife identification and localization based on oblique photogrammetry. The method includes: acquiring oblique photogrammetric data for a target area, the oblique photogrammetric data including multiple multi-view aerial images and corresponding position and attitude data for each multi-view aerial image, wherein each multi-view aerial image contains images acquired from multiple different directions at the same location; constructing a 3D model of the target area based on the oblique photogrammetric data, and generating a true orographic image of the target area based on the 3D model; performing wildlife identification based on each multi-view aerial image to obtain animal identification results; and, based on the animal identification results and the position and attitude data, performing wildlife localization annotation on the true orographic image to generate a wildlife distribution map of the target area. This method can provide intuitive and accurate data support for population monitoring, habitat assessment, and ecological protection decision-making.

Description

Technical Field

[0001] This invention relates to the field of aerial photogrammetry, and in particular to a method and system for identifying and locating wild animals based on oblique photogrammetry. Background Technology

[0002] The population size and spatial distribution of large wild animals are crucial basic information for ecological conservation. Currently, monitoring is mainly carried out using ground-based methods such as manual patrols, transect surveys, and infrared cameras. However, these methods are limited by factors such as complex terrain, harsh climate, and poor accessibility for personnel, resulting in limited coverage. Furthermore, ground activities can easily disturb wild animals.

[0003] In recent years, drone aerial photogrammetry has begun to be applied to wildlife monitoring. However, existing technologies mostly rely on orthophotos from a single vertical perspective for planar interpretation. It is difficult to use the three-dimensional morphological characteristics of animals to distinguish them from interfering ground features such as rocks and bushes, resulting in low identification accuracy. At the same time, it is also impossible to obtain high-precision three-dimensional coordinates of individual animals, leading to large positioning errors. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for identifying and locating wild animals based on oblique photogrammetry, so as to solve the above-mentioned technical problem.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A method for wildlife identification and localization based on oblique photogrammetry, comprising: acquiring oblique photogrammetric data for a target area, wherein the oblique photogrammetric data includes multiple multi-view aerial images and position and attitude data corresponding to each multi-view aerial image, wherein each multi-view aerial image contains images acquired from multiple different directions for the same location; constructing a three-dimensional model of the target area based on the oblique photogrammetric data, and generating a true orthophoto of the target area based on the three-dimensional model; performing wildlife identification based on each multi-view aerial image to obtain animal identification results; and performing wildlife localization annotation on the true orthophoto based on the animal identification results and each position and attitude data to generate a wildlife distribution map of the target area.

[0006] The beneficial effects of this invention are as follows: This invention acquires oblique photographic data containing multi-directional images of the same location, constructs a three-dimensional model and generates a true orographic image that eliminates terrain distortion, then performs three-dimensional stereoscopic identification of wild animals, and combines position and posture data to perform precise positioning and annotation on the true orographic image to achieve accurate spatial positioning, and finally generates a wild animal distribution map of the target area, which can provide intuitive and accurate data support for population monitoring, habitat assessment and ecological protection decision-making.

[0007] Based on the above technical solution, the present invention can be further improved as follows.

[0008] Furthermore, the step of constructing a three-dimensional model of the target area based on the oblique photography data includes: preprocessing the various multi-view aerial images to obtain preprocessed images; wherein the preprocessing includes at least color consistency correction and geometric distortion correction; combining the position and attitude data of each location, performing aerial triangulation on the preprocessed images to generate a densified point cloud; and constructing a three-dimensional model of the target area based on the densified point cloud using a block modeling approach.

[0009] Furthermore, the step of combining the attitude data of each position to perform aerial triangulation on the preprocessed image to generate a densified point cloud includes: performing multi-view joint adjustment on the preprocessed image, using a preset pyramid matching strategy to match corresponding points, and performing iterative calculation of regional network adjustment in combination with the attitude data of each position to generate a densified point cloud.

[0010] Furthermore, the step of constructing a 3D model of the target region based on the encrypted point cloud using a block-based modeling approach includes: dividing the target region into multiple sub-regions with the center point as the origin; performing dense point cloud matching on each sub-region based on the encrypted point cloud to generate a high-density point cloud corresponding to each sub-region; converting the high-density point cloud of each sub-region into an irregular triangular mesh model and assigning texture to it to obtain a 3D model of each sub-region; and stitching together the 3D models of each sub-region to obtain a 3D model of the target region.

[0011] Furthermore, the method also includes: selecting image pairs that meet preset overlap conditions from various multi-view aerial images as stereo image pairs; constructing a three-dimensional stereo frame range for each wildlife community area within the target area based on each stereo image pair; for each wildlife community area, counting the wildlife within the three-dimensional stereo frame range of the wildlife community area according to a preset animal size threshold to obtain an initial count of the wildlife community area; for each wildlife community area, dividing the true orthophoto image into multiple grids, counting the total number of pixels belonging to wildlife in each grid, and calculating the number of animals based on the total number of pixels belonging to wildlife in each grid and a pre-calibrated average number of pixels per animal to obtain a secondary count of the wildlife community area; for each wildlife community area, comparing the difference between the initial count and the secondary count to determine the final number of animals in the wildlife community area based on the degree of difference; and integrating the final number of animals in each wildlife community area as attribute information into the wildlife distribution map.

[0012] Furthermore, for each wildlife colony area, the difference between the initial count and the secondary count of the wildlife colony area is compared to determine the final animal count of the wildlife colony area based on the degree of difference. This includes: calculating the difference between the initial count and the secondary count of the wildlife colony area; when the difference is less than or equal to a first preset threshold, calculating the average of the initial count and the secondary count of the wildlife colony area as the final animal count of the wildlife colony area; when the difference is greater than the first preset threshold and less than or equal to a second preset threshold, obtaining frame-by-frame verification results to correct the count based on the frame-by-frame verification results and determine the final animal count of the wildlife colony area; wherein, the frame-by-frame verification results represent the animal count data obtained after performing a frame-by-frame verification of stereo image pairs on the wildlife colony area; when the difference is greater than the second preset threshold, re-performing a three-dimensional stereo frame selection count on the wildlife colony area to determine the final animal count of the wildlife colony area based on the obtained count results.

[0013] Furthermore, the method also includes: obtaining the actual geographical area of ​​each wildlife community; calculating the population density of the wildlife community based on the final animal population and the actual geographical area of ​​the wildlife community; and incorporating the population density of each wildlife community as attribute information into the wildlife distribution map.

[0014] To address the aforementioned technical problems, the present invention also provides a wildlife identification and positioning system based on oblique photogrammetry, comprising: The data acquisition module is used to acquire oblique photography data for the target area. The oblique photography data includes multiple multi-view aerial images and position and attitude data corresponding to each multi-view aerial image. Each multi-view aerial image contains images acquired from multiple different directions for the same location. An image processing module is used to construct a three-dimensional model of the target area based on the oblique photogrammetry data, and to generate a true radiometric image of the target area based on the three-dimensional model. The animal identification module is used to identify wild animals based on aerial images from various multi-view perspectives and obtain animal identification results. The distribution map generation module is used to perform wildlife location annotation on the true radio image based on the animal identification results and the posture data of each location, so as to generate a wildlife distribution map of the target area.

[0015] To address the aforementioned technical problems, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the wildlife identification and localization method based on oblique photogrammetry as described above.

[0016] To address the aforementioned technical problems, the present invention also provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the previously described method for wildlife identification and localization based on oblique photogrammetry. Attached Figure Description

[0017] Figure 1 This is a flowchart of the wildlife identification and localization method based on oblique photogrammetry according to the present invention; Figure 2 This is a schematic diagram of the sub-region division of the present invention; Figure 3 This is a schematic diagram of a high-density point cloud corresponding to a sub-region of the present invention. Figure 4 This is a schematic diagram of the irregular triangular mesh model generation process of the present invention; Figure 5 This is a schematic diagram of a three-dimensional model of the present invention; Figure 6 This is a schematic diagram of the stereoscopic interpretation process of the present invention; Figure 7 This is a schematic diagram of the distribution of wild animals according to the present invention; Figure 8 This is a schematic diagram of the wildlife identification and positioning system based on oblique photogrammetry according to the present invention; Figure 9 This is a schematic diagram of the electronic device of the present invention. Detailed Implementation

[0018] The principles and features of the present invention are described below. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0019] As mentioned earlier, the population size, spatial distribution, and changes of large wild animals are crucial foundational information for ecosystem assessment, biodiversity conservation, and natural resource management. In natural regions such as plateaus, deserts, and meadows, wild animals have wide ranges and complex habitats, making continuous and accurate monitoring data essential for ecological conservation decision-making and scientific research. However, these regions typically feature varied terrain, complex climate conditions, and high costs associated with human interference, placing higher demands on monitoring technologies in terms of coverage, timeliness, and operational safety.

[0020] Currently, large wildlife monitoring primarily relies on manual patrols, transect surveys, and the deployment of infrared cameras to obtain animal activity information. With the development of remote sensing technology, methods utilizing aircraft or drones to acquire high-resolution imagery and interpret surface information are gradually being applied. In the field of aerial photogrammetry, existing technologies mostly acquire surface image data through vertically downward-looking images and perform aerial triangulation and orthorectification to generate orthophotos with a unified spatial reference for topographic mapping and feature identification. To enhance the three-dimensional representation of features, oblique photogrammetry, by acquiring multi-view images from multiple directions, reconstructs the side views and three-dimensional structures of features and is now widely used in urban modeling and terrain reconstruction.

[0021] While the aforementioned technologies have made some progress in surface information acquisition and 3D modeling, they still have shortcomings when applied to the refined monitoring of large wildlife. On the one hand, manual patrols and ground surveys are limited by the number of personnel and environmental conditions, resulting in limited coverage and making it difficult to achieve continuous monitoring of large-scale areas. On the other hand, monitoring methods based on single-vertical-view aerial imagery are easily affected by occlusion and background interference in complex terrain conditions, leading to low accuracy in identifying small animals or those that blend into their surroundings. Furthermore, existing oblique photogrammetry data processing workflows primarily focus on 3D model construction, lacking refined color homogenization and geometric distortion correction in the aerial data preprocessing stage. Aerial triangulation does not fully consider the occlusion relationships between multi-view images, and 3D model construction often employs a holistic processing approach, resulting in high computational resource consumption and low processing efficiency under large-scale operational conditions. In the animal identification stage, existing technologies largely rely on single-image interpretation, failing to fully utilize the stereo image pairs formed under high image overlap conditions, making it difficult to accurately identify individual animal targets.

[0022] Furthermore, in existing technologies, animal target identification results are mostly in the form of image annotations or independent data, lacking effective integration with true orthophoto results or geographic information systems. Animal location annotation and quantity statistics are separated from a unified spatial reference system, making it difficult to intuitively reflect the spatial distribution characteristics of animals, thus limiting their application value in ecological protection assessment and management decision-making.

[0023] In summary, existing technologies for monitoring large wild animals using aerial imagery still suffer from problems such as insufficient identification accuracy, inaccurate spatial positioning, and a disconnect between image processing workflows and distribution statistics applications. There is still a lack of a technical solution that can fully utilize multi-view oblique photography images to achieve automatic identification, accurate positioning, and generation of spatial distribution statistics results for large wild animals.

[0024] Example 1 Based on this, such as Figures 1 to 7As shown, this embodiment provides a method for wildlife identification and localization based on oblique photogrammetry, including: S101. Acquire oblique photography data for the target area. The oblique photography data includes multiple multi-view aerial images and position and attitude data corresponding to each multi-view aerial image. Each multi-view aerial image contains images acquired from multiple different directions for the same location.

[0025] S102. Construct a three-dimensional model of the target area based on the oblique photogrammetry data, and generate a true radiometric image of the target area based on the three-dimensional model.

[0026] S103. Based on various multi-view aerial images, wildlife identification is performed to obtain animal identification results.

[0027] S104. Based on the animal identification results and the posture data of each location, perform wildlife location annotation on the true radio image to generate a wildlife distribution map of the target area.

[0028] Specifically, the oblique photography data is acquired using an oblique photography system mounted on a drone. Each multi-view aerial image contains images from different angles captured simultaneously by six cameras in the oblique photography system. The position and attitude data corresponding to each multi-view aerial image mainly consists of precise position information obtained by the POS system.

[0029] Optionally, in an embodiment, constructing a three-dimensional model of the target area based on oblique photography data includes: preprocessing various multi-view aerial images to obtain preprocessed images; wherein, the preprocessing includes at least color consistency correction and geometric distortion correction; combining the attitude data of various positions, performing aerial triangulation on the preprocessed images to generate a densified point cloud; and constructing a three-dimensional model of the target area based on the densified point cloud using a block-based modeling approach.

[0030] First, the acquired oblique photogrammetry data undergoes an integrity check, discarding blurry or missing invalid images. The remaining valid images are then subjected to color homogenization processing to ensure color consistency across images acquired from different viewpoints and at different times. Simultaneously, distortion correction is performed to eliminate geometric distortions caused by camera lens and shooting angle, ensuring the geometric accuracy of the images and laying the data foundation for subsequent processing.

[0031] Color homogenization can be achieved by combining histogram equalization with color normalization algorithms to eliminate color deviations caused by lighting and shooting angle. Distortion correction targets the radial and tangential distortions of the camera lens to ensure that the geometric deviation of the corrected image is ≤0.5 pixels.

[0032] Optionally, in an embodiment, aerial triangulation is performed on the preprocessed image by combining the attitude data of each position to generate an encrypted point cloud. This includes: performing multi-view joint adjustment on the preprocessed image, matching corresponding points using a preset pyramid matching strategy, and performing iterative calculation of regional network adjustment by combining the attitude data of each position to generate an encrypted point cloud.

[0033] In this embodiment, multi-view joint adjustment is implemented based on Photomesh software. Vertical downward-looking images and oblique images are mixed for adjustment. The surface projection range of all images is simulated by combining the exterior orientation elements of the POS system and the camera installation position. A coarse-to-fine pyramid matching strategy is adopted to automatically match corresponding points on each level of imagery, completing the free-network bundle adjustment. Connector point and control point coordinate files are established, and iterative calculations for regional network adjustment are performed using GPS / IMU information until the geometric accuracy of the aerial photography data meets the requirements for 1:500 topographic mapping and 3D modeling. Finally, an aerial triangulation accuracy report is output. Specifically, the pyramid matching strategy layers the images at 1 / 2, 1 / 4, and 1 / 8 scales, performing coarse matching followed by fine matching. The matching algorithm uses Scale-Invariant Feature Transform (SIFT) feature point matching, with a corresponding point search radius of 10 pixels.

[0034] Optionally, in this embodiment, based on the encrypted point cloud, a block-based modeling approach is used to construct a 3D model of the target region, including: dividing the target region into multiple sub-regions with the center point as the origin; performing dense point cloud matching on each sub-region based on the encrypted point cloud to generate a high-density point cloud corresponding to each sub-region; converting the high-density point cloud of each sub-region into an irregular triangular mesh model and assigning texture to it to obtain a 3D model of each sub-region; and stitching together the 3D models of each sub-region to obtain a 3D model of the target region.

[0035] In this embodiment, to address the computational resource consumption problem of large work areas, a block-based computing strategy is adopted, automatically dividing the modeling area with the center point of the target area as the origin, such as... Figure 2 As shown; point cloud matching is performed on the stereo image pairs within each block to generate a high-density point cloud, such as... Figure 3 As shown, the high-density point cloud is converted into an irregular triangular network (TIN) model, and the model is smoothed and optimized to eliminate defects such as spikes and holes. Figure 4 As shown; based on the spatial location of the TIN model, texture information is obtained from the image at the best viewing angle, automatically assigned to the TIN model, and outputs a fully textured 3D model, as shown. Figure 5As shown in the figure. The point cloud matching under the block-based computation strategy employs a gray-scale-based semi-global matching algorithm, with a point cloud density set to 10 points / square centimeter. The TIN model is constructed using the Delaunay triangulation algorithm after filtering and denoising, and a Gaussian smoothing algorithm is used to optimize the model, eliminating burrs and voids.

[0036] Furthermore, an iterative modification process of verification-editing-re-verification was established. The 3D model was meticulously verified in multiple scenes from a 200-meter perspective. The verification content included the integrity of building textures, the flatness of water bodies, the integrity of roads and vehicles, and the realism of terrain. For unqualified areas found during the verification, the point cloud, TIN model, and texture were edited using the editing module of Photomesh software. The edited model was verified again until the model's accuracy and integrity met the requirements. Finally, all model results were systematically organized, the file naming and storage formats were standardized, and documentation was attached.

[0037] The verified and modified 3D model is converted into a true orthophoto image with geographic coordinate information, ensuring the geographic coordinate accuracy and visual quality of the image, and providing an accurate image base for wildlife identification and location. In this embodiment, the true orthophoto image is generated using a pixel-by-pixel correction method based on the 3D model, with an output resolution of 0.5 meters / pixel, a projection coordinate system of WGS84 / UTM Zone 45N, and a geographic coordinate accuracy error ≤0.5 meters.

[0038] In this embodiment, the preset overlap condition is set to an overlap rate > 65%. Based on 5cm-level high-resolution original images (the design value for the Sanjiangyuan field test flight is 5cm, and the actual measured value is 7.0cm), stereo image pairs with an overlap rate > 65% are used for stereo interpretation to accurately identify groups or single large wild animals (such as Tibetan antelopes and yaks). Figure 6 As shown. Sample images containing wild animals were extracted and uniformly cropped to a standard size of 512×512 pixels. The animals were then classified based on features such as morphology, color, and body size. Combined with precise location information from the POS system, the wild animals were accurately located and labeled on the true orthophotos. The labeled image data was then imported into the ArcGIS geographic information system, where the species and quantity of wild animals were identified, generating a large-scale wild animal location distribution map, such as... Figure 7 As shown, statistical analysis of population distribution was completed, providing data support for ecological research and wildlife conservation planning.

[0039] Optionally, in the embodiments, the method further includes: selecting image pairs that meet preset overlap conditions from various multi-view aerial images as stereo image pairs; constructing a three-dimensional stereo frame range for each wildlife community area within the target area based on each stereo image pair; for each wildlife community area, counting the wildlife within the three-dimensional stereo frame range of the wildlife community area according to a preset animal size threshold to obtain an initial count of the wildlife community area; for each wildlife community area, dividing the true orthophoto into multiple grids, counting the total number of pixels belonging to wildlife in each grid, calculating the number of animals based on the total number of pixels belonging to wildlife in each grid and a pre-calibrated average number of pixels per animal to obtain a secondary count of the wildlife community area; for each wildlife community area, comparing the difference between the initial count and the secondary count to determine the final number of animals in the wildlife community area based on the degree of difference; and integrating the final number of animals in each wildlife community area as attribute information into the wildlife distribution map.

[0040] Optionally, in an embodiment, for each wildlife colony area, the difference between the initial count and the secondary count of the wildlife colony area is compared to determine the final animal count of the wildlife colony area based on the degree of difference. This includes: calculating the difference between the initial count and the secondary count of the wildlife colony area; when the difference is less than or equal to a first preset threshold, calculating the average of the initial count and the secondary count as the final animal count of the wildlife colony area; when the difference is greater than the first preset threshold and less than or equal to a second preset threshold, obtaining frame-by-frame verification results to correct the count based on the frame-by-frame verification results and determine the final animal count of the wildlife colony area; wherein, the frame-by-frame verification results represent the animal count data obtained after performing a frame-by-frame verification of stereo image pairs on the wildlife colony area; when the difference is greater than the second preset threshold, re-counting the wildlife colony area using a three-dimensional stereo frame selection to determine the final animal count of the wildlife colony area based on the obtained count results.

[0041] Based on the herding characteristics of large wild animals (scattered groups, dense groups, mixed groups), this method combines information from oblique stereo image pairs and true orthophotos, and employs a three-level counting method of stereo frame selection + pixel calibration + manual verification to avoid counting errors from single images. It is applicable to counting animals of different sizes, such as Tibetan antelopes, yaks, and Tibetan wild asses.

[0042] Level 1 (Initial 3D bounding box selection): Based on stereo image pairs with an overlap rate > 65%, a 3D bounding box is constructed in Photomesh software to define the area of ​​wildlife populations. 3D recognition thresholds are set according to animal size (taking the core species of the Three-River-Source region as an example, the thresholds are set as follows: Tibetan antelope height 0.8-1.2m, length 1.2-1.5m; yak height 1.4-1.8m, length 2-3m). Interference objects such as rocks and shrubs within the bounding box are removed. Animal targets within the bounding box are automatically counted to obtain the initial number N1.

[0043] Secondary step (pixel calibration correction): The pixel scale of the true orthophoto is calibrated with the actual terrain of the target area (1 pixel = 0.5m, matching the resolution of the true orthophoto). The herd area is divided into grids (grid size 5m × 5m). Within each grid, a secondary count is performed based on the pixel proportion of the animal (approximately 80-120 pixels per Tibetan antelope and approximately 300-500 pixels per yak), resulting in the secondary count N2.

[0044] Furthermore, for dense herds (such as Tibetan antelope migratory herds), a pixel density method is used for correction: the number of secondary counts N2 = the total number of animal pixels in the grid / the average number of pixels per animal, while deducting overlapping pixels (the overlapping pixel correction coefficient is 0.92-0.95, based on field sample statistics).

[0045] Level 3 (Manual Verification): The difference between N1 and N2 is verified in stages. In this embodiment, the first preset threshold is set to 5%, and the second preset threshold is set to 15%. When the difference is ≤5%, the average value is taken as the final number N; when 5% < difference ≤15%, the difference area is checked frame by frame using stereo image pairs to correct the count; when the difference is >15%, stereo selection is performed again to exclude omissions / overcounts caused by terrain occlusion (such as river valley depressions, shrub areas, etc.) and finally the accurate number N is determined.

[0046] Optionally, in an embodiment, the method further includes: obtaining the actual geographical area of ​​each wildlife community; calculating the population density of the wildlife community based on the final animal population and the actual geographical area of ​​the wildlife community; and incorporating the population density of each wildlife community as attribute information into the wildlife distribution map.

[0047] Specifically, using ArcGIS's spatial analysis module, the actual geographic area of ​​animal population clusters (excluding invalid areas obscured by terrain) is extracted, and population density is calculated. D=N / S; Where D is the population density (unit: individuals / head / km²) 2 N represents the final number of animals; S represents the actual geographical area.

[0048] In operations in high-altitude areas (such as the Three Rivers Source Region), area calculations need to take into account the slope correction of the plateau terrain. In this embodiment, when the slope is greater than 25°, the area correction factor is taken as 1.15.

[0049] In summary, this method has the following significant advantages compared to existing technologies: 1. Improve the accuracy of aerial photography data processing: By using multi-view joint adjustment and a coarse-to-fine pyramid matching strategy, the geometric deformation and occlusion relationship between images are taken into account, which improves the accuracy of aerial triangulation by more than 20%. The Sanjiangyuan field adjustment was iterated 4 times, with a measured mean square error of 0.14m and a measured mean square error of 0.19m. The matching rate of corresponding points was 98.5%, which significantly improved the matching rate of corresponding points and reduced occlusion errors. 2. Optimize processing efficiency for large-scale work areas: Adopting a block-based calculation strategy with the center point of the work area as the origin, the computational resource consumption for 3D model construction is reduced by 25%, and in the Sanjiangyuan operation, the computational resource consumption is reduced by 28%, while the model construction efficiency is improved by 30%. This is suitable for areas >100km². 2 A large aerial photography operation area; 3. Improve the accuracy of wildlife identification: By using stereo image pairs with an overlap rate of >65% for stereo interpretation, the limitations of single-image interpretation are overcome, enabling wildlife identification accuracy to reach over 95%, with an accuracy rate of 96.2% in the Sanjiangyuan operation; 4. Achieving Precise Positioning and a Closed Loop for Ecological Application: By combining the location information of the POS system with the ArcGIS geographic information system, precise location of wild animals is achieved, and population distribution statistics are generated. The oblique photogrammetry data processing and wild animal ecological monitoring are deeply integrated, forming a closed loop from data collection, processing to application, which improves the efficiency of wildlife conservation decision-making. In the application of this invention in the core protected area of ​​Sanjiangyuan, compared with the traditional monitoring method of infrared cameras + manual patrols, the monitoring coverage is increased by 10 times, the data collection time is shortened by 80%, and the error in animal population statistics is reduced from 30% to less than 5%, which fully meets the needs of refined wildlife monitoring in Sanjiangyuan National Park. 5. Provides non-invasive monitoring methods: This invention uses aerial photography data to identify and locate wild animals without the need for ground inspections, thus avoiding interference with the habitat of wild animals. It provides a non-invasive technical means for monitoring large wild animals in complex terrains such as plateaus and mountains, and can be widely used in biodiversity assessment and natural resource management.

[0050] Example 2 This embodiment uses the Tibetan antelope-yak core protected area in Qumalai County, Yushu Prefecture, Sanjiangyuan National Park (95°12′-95°48′E, 34°20′-34°50′N) as the aerial photography operation area, covering 60 km². 2The region's terrain is characterized by alternating plateaus, hills, and valleys, making it a complex area where traditional monitoring methods struggle to achieve high-precision, non-invasive wildlife population statistics. The system of this invention serves as the dedicated hardware and software platform for implementing the method. All operational steps of the method are completed collaboratively by the system's modules, achieving a closed-loop process from aerial data acquisition to wildlife conservation data output.

[0051] The system of this invention consists of an unmanned aerial vehicle platform, an oblique photography system, a data processing system, and a ground control system. Each module achieves real-time data interaction and command transmission through a wireless data transmission radio + 4G / 5G dual-mode communication. The hardware and software are precisely adapted, and all software functional modules are developed based on the raw data collected by the hardware. The specific configuration is as follows.

[0052] Unmanned Aerial Vehicle (UAV) Platform: In this embodiment, the DM150 fixed-wing UAV with long endurance for high-altitude applications is selected, equipped with a POS system. The airframe is made of T700 all-carbon fiber composite material in a single piece, with a wingspan of 400cm, a fuselage length of 226cm, a height of 57cm, a standard takeoff weight of 25kg, and a maximum payload of 9kg. The power system is equipped with a high-altitude-specific electronic fuel injection engine, matched with an 18Ah high-capacity lithium battery. The high-altitude-optimized power system can adapt to flight environments at an altitude of 4500 meters, with a service ceiling of 6500 meters. In field test flights in the Yushu area of ​​the Sanjiangyuan region, the actual endurance at an altitude of 4500m was 10.5 hours, with a wind resistance level of 7, capable of withstanding the sudden strong winds of spring in the Sanjiangyuan region, and enabling high-resolution aerial data acquisition in the complex terrain of the Sanjiangyuan region. The airborne POS system is a high-precision integrated navigation POS (Position and Orientation System) / IMU (Inertial Measurement Unit) system with a positioning accuracy of ≤0.1m and an attitude measurement accuracy of ≤0.01°. It can collect the UAV's position, speed, attitude and other external orientation elements in real time and transmit them synchronously to the ground control system and data processing system, providing a precise geographic coordinate basis for subsequent wildlife positioning.

[0053] The system employs a KG661 ultra-wide-angle six-camera configuration, comprising three cameras in the front group and three cameras in the rear group, enabling simultaneous acquisition of vertical downward-looking and multi-angle tilted images. Specifically, the middle camera in the front group is tilted forward at 30°, the left camera is tilted forward at 30° and tilted to the left at 29°, and the right camera is tilted forward at 30° and tilted to the right at 29°. Similarly, the middle camera in the rear group is tilted backward at 30°, the left camera is tilted backward at 30° and tilted to the left at 29°, and the right camera is tilted backward at 30° and tilted to the right at 29°.

[0054] At an altitude of 1000 meters, a single shot covers a length of 1860 meters and a width of 475 meters. At the Sanjiangyuan region, at 1000 meters altitude, the actual shooting coverage area is 1850 meters long and 470 meters wide, with a measured forward overlap of 79% and a measured lateral overlap of 66%, and a ground resolution of 7.0 cm. This represents a coverage area increase of over 30% compared to traditional 5-lens systems, while maintaining a forward overlap of ≥78%, a lateral overlap of ≥65%, and a ground resolution of over 5 cm, suitable for identifying small and large animals such as Tibetan antelopes. All cameras have 50-megapixel resolution, a 35mm lens focal length, a maximum shooting frame rate of 5fps, and are equipped with standardized lens hoods and image-stabilized gimbals, allowing for simultaneous triggering of shooting. This represents a 32% increase in coverage area compared to traditional 5-lens systems, while ensuring images are free of motion blur and overexposure, meeting high-resolution processing requirements.

[0055] The ground control system is used for UAV flight path planning and aerial photography parameter (flight altitude, speed, and shooting frequency) setting. It also enables real-time monitoring of UAV flight status and aerial photography data acquisition, and can adjust aerial photography strategies according to the actual monitoring needs of the Sanjiangyuan region. The ground control system mainly consists of a UAV flight control station, a data receiving terminal, and a large display screen, equipped with professional flight control software and aerial photography data preprocessing and preview software. The flight control software supports manual / automatic dual-mode flight path planning, can generate contour lines based on the terrain of the work area, and supports real-time adjustment of aerial photography parameters. It can operate normally in extreme environments with an altitude of 5000 meters and temperatures ranging from -20℃ to 50℃, meeting the needs of field operations in the Sanjiangyuan region. The data receiving terminal can simultaneously receive UAV POS data and image previews from the oblique photography system, while the large display screen can display the UAV flight status, aerial photography coverage, and data acquisition progress in real time.

[0056] The data processing system is a computer processing terminal equipped with professional software. It utilizes an industrial-grade workstation with an Intel i9-14900K CPU, 128GB DDR5 memory, an RTX4090 24GB professional graphics card, and a 20TB enterprise-grade solid-state drive array, supporting multi-threaded parallel computing. It is equipped with Photomesh 2024 Professional Edition (including aerial triangulation, 3D modeling, and model editing modules) and ArcGIS 10.8 Geographic Information System (including spatial analysis, feature annotation, and distribution map generation modules). Photomesh is used for aerial data preprocessing, aerial triangulation, 3D model construction and modification, and true orthophoto output. ArcGIS is used for wildlife location annotation, species and quantity identification, and distribution map generation. The data processing system comes pre-installed with a wildlife image recognition and sample extraction plugin. This plugin uses a SIFT+Convolutional Neural Network (CNN) fusion recognition algorithm based on the characteristics of animals from the Three-River-Source region, enabling seamless integration with Photomesh and ArcGIS to complete standardized sample cropping and feature classification. The data processing system is equipped with a gigabit wired network port and a wireless receiving module, enabling both real-time reception and offline batch processing of aerial photography data.

[0057] The SIFT+CNN fusion recognition algorithm works as follows: First, leveraging the rotation and scale-invariant properties of the SIFT algorithm, candidate regions for animals are quickly and robustly detected and located in oblique images. Then, these candidate regions are fed into a pre-trained convolutional neural network (CNN) model. By simulating the hierarchical structure of human vision, the CNN automatically learns and extracts deeper features such as texture, shape, and color of the animals to complete the final species classification.

[0058] The pre-trained convolutional neural network (CNN) model is obtained as follows: A large database of labeled wildlife images is prepared. An animal image (such as a photo of a Tibetan antelope) is input into the initial CNN model, and the network provides a prediction. The prediction is compared with the image's true label, and a loss function calculates the difference between the two. Based on the calculated loss, the model's parameters are adjusted. Ultimately, by learning from a massive amount of data, the CNN continuously improves its predictions, making them increasingly closer to the true answer.

[0059] In response to the terrain features of the operation area in Qumalai County, Sanjiangyuan, and the monitoring needs of Tibetan antelopes and yaks, the ground control system was used to complete the setting of all parameters for aerial photography operations. All parameters were designed to adapt to the processing requirements of subsequent methods and steps, ensuring that the collected aerial photography data can be directly used for operations such as aerial triangulation and stereo interpretation without the need for secondary format conversion or parameter correction. The specific settings are as follows.

[0060] Flight parameters: Parallel flight path planning is adopted, with a flight altitude of 1200 meters, a relative flight altitude of 1150 meters, and a ground resolution of 7.5 cm (7.2 cm measured in the Sanjiangyuan field test flight); the flight path direction is consistent with the main terrain of the operation area, the flight path spacing is 400 meters, a total of 72 flight paths are designed, the length of a single flight path is 55 kilometers, and the total flight distance is 3960 kilometers; the flight speed is 80 km / h, and the aerial photography adopts a fully automatic flight mode with constant altitude, constant speed, and constant heading.

[0061] Shooting parameters: Six cameras simultaneously triggered shooting using an oblique photography system, shooting frequency 2fps, shutter speed 1 / 1000s, aperture F8, ISO 100, and automatic white balance calibration; the forward overlap was set to 78% and the lateral overlap to 65%, while the actual measured forward overlap during the Sanjiangyuan field test flight was 79% and the lateral overlap was 66%, meeting the core requirements of stereo pair construction and corresponding point matching in the method; the image storage format was RAW+JPG dual format, with RAW format used for subsequent professional processing and JPG format used for real-time preview.

[0062] Data transmission parameters: Flight commands are transmitted between the UAV platform and the ground control system via a 433MHz wireless data transmission radio with a transmission distance of ≥20km; Image data and POS data from the oblique photography system are transmitted to the data processing system in real time via 5G dual-mode communication with a real-time transmission bandwidth of ≥100Mbps, and offline data is stored on a 2T onboard solid-state drive to ensure no data loss.

[0063] In this embodiment, with the above parameter settings, the entire aerial photography operation lasted 36 hours, collecting 142,000 valid aerial images and 142,000 POS data entries, with an image data volume of approximately 890G. 23 blurry / overexposed images were removed (overexposure due to strong radiation in the Sanjiangyuan region accounted for 0.016%). All data met the processing standards for subsequent method steps.

[0064] First, data preprocessing is performed by the data processing system, which includes the following steps: Integrity check: Photomesh software's batch processing function automatically removes blurry, overexposed, underexposed, and missing invalid images. In this example, 23 blurry / overexposed images were removed, retaining 141,977 valid images. Simultaneously, the POS data and image capture timestamps are verified, and POS data without corresponding images is deleted to ensure a one-to-one match.

[0065] Color uniformity processing: Histogram equalization combined with color normalization algorithm is used to correct the color of images collected at different times and from different angles, eliminating color deviations caused by strong ultraviolet rays, lighting and shooting angle, so that the color mean and variance of all images are consistent, ensuring the coordination of subsequent texture stitching.

[0066] Distortion correction: Based on the interior orientation element calibration parameters of the 6-camera oblique photogrammetry system, the radial and tangential distortion of the image is corrected through the camera correction module of Photomesh software to eliminate the geometric distortion caused by the lens. The geometric deviation of the corrected image is ≤0.5 pixels.

[0067] Data integration: The corrected images and matching POS data are classified according to flight strips to generate standardized aerial photography data file packages, which are stored in a format that Photomesh software can directly recognize, thus preparing data for subsequent aerial triangulation encryption.

[0068] Aerial triangulation encryption is performed using Photomesh software based on the data processing system. Based on the pre-processed aerial photography data, a multi-view joint adjustment strategy is adopted to mix and adjust vertical downward-looking images with multi-angle oblique images. This primarily addresses the occlusion problem in the complex terrain of the Three-River-Source region, improving the matching rate of corresponding points. The aerial triangulation encryption mainly includes the following specific operations: Projection range simulation: Import the camera interior orientation elements and the POS system exterior orientation elements, simulate the surface projection range of all images from 6 cameras in Photomesh software, mark the terrain occlusion areas of the Three Rivers Source Region, and exclude invalid areas for matching corresponding points.

[0069] Pyramid matching: A coarse-to-fine pyramid matching strategy is adopted, dividing the image into layers at 1 / 2, 1 / 4, and 1 / 8 ratios. First, coarse matching is performed on the low-resolution image to obtain the initial positions of corresponding points, and then fine matching is performed on the high-resolution image to improve matching accuracy and efficiency. The matching algorithm uses SIFT feature point matching, with a corresponding point search radius of 10 pixels.

[0070] Adjustment calculation: First, free net bundle adjustment is performed on the matched corresponding points to remove outliers. The outlier removal threshold is 3 times the mean square error. Then, combined with the ground control points of the work area (a total of 12 high-precision GPS control points are set up, with a plane accuracy of ≤0.05m), iterative calculation of regional network adjustment is carried out. In this embodiment, a total of 4 iterations are performed until the geometric accuracy of the aerial photography data meets the requirements of 1:500 topographic mapping and 3D modeling.

[0071] Output: Output aerial triangulation results, including coordinate files of connection points and control points, correction values ​​of image exterior orientation elements, and generate an aerial triangulation accuracy report. In this embodiment, the plane error of the aerial triangulation is ≤0.14m, the elevation error is ≤0.19m, and the matching rate of corresponding points is improved by 22.5% to 98.5%, which meets the requirements of subsequent 3D modeling.

[0072] A 3D model was constructed using Photomesh software, a data processing system. This was applied to a 60km... 2In the large-scale operation area of ​​the Three-River-Source region, a block-based computing strategy was adopted. Using the geographical center of the operation area as the origin, the area was divided into 12 rectangular modeling sub-regions of 5km × 1km. The 3D model was constructed block by block and then stitched together as a whole, effectively reducing computational resource consumption. The 3D model construction mainly includes the following specific operations: Block point cloud matching: Dense point cloud matching is performed on stereo image pairs within each block. A gray-scale-based semi-global matching algorithm is adopted, and the point cloud density is set to 10 points / square centimeter. The measured density in the Sanjiangyuan alpine meadow area is 9.8 points / square centimeter. High-density point clouds are generated for each block to ensure that the point clouds can clearly reflect the topographic relief and land feature characteristics.

[0073] TIN model construction: The high-density point cloud of each block is filtered and denoised to remove invalid point clouds such as vegetation and birds. Then, the point cloud is converted into an irregular triangular network model by Delaunay triangulation algorithm, and the model is optimized by Gaussian smoothing algorithm. The smoothing iteration is 6 times in the Sanjiangyuan River Valley wetland area and 5 times in other areas to eliminate defects such as burrs and voids in the model.

[0074] Texture mapping: Based on the spatial location and normal vector of the TIN model, the best aerial photograph from the best perspective is automatically selected as the texture source in Photomesh software. The texture stretching, stitching and fusion algorithms are used to accurately assign the texture to the TIN model, ensuring the continuity and realism of the model texture, with no obvious stitching seams. The model texture restoration accuracy reached 95% in the Sanjiangyuan operation.

[0075] Model stitching: The 12 segmented 3D models are stitched together according to geographic coordinates to generate a 60km model. 2 A complete 3D model of the work area, in OSGB format, supporting multi-scale browsing and editing.

[0076] In this embodiment, block-based computing reduces the computational resource consumption for 3D model construction by 28%, and the overall modeling time is 52 hours, which is 33% more efficient than the traditional overall modeling method.

[0077] The 3D model was modified using Photomesh software, a data processing system. An iterative modification process of verification-editing-re-verification was established. The 3D model was verified across the entire scene from a 200-meter viewing height (consistent with the commonly used viewing angle for subsequent wildlife identification) to ensure that the model's accuracy and integrity met the requirements for true orthophoto output and wildlife localization. The specific operations are as follows: Multi-scene verification: In the 3D browsing module of Photomesh software, the model was verified region by region from a viewing height of 200 meters. The core verification contents included: the integrity of building textures, the smoothness of water surfaces, the integrity of road and vehicle outlines, and the realism of terrain. In this example, four river valley water areas were found to have insufficient smoothness and three hilly terrain textures were found to be missing, which are unqualified areas of the model.

[0078] Targeted editing: Using the editing module of Photomesh software, substandard areas were precisely edited: For areas with insufficient water surface smoothness, point cloud smoothing combined with texture remapping was used for correction. For areas with missing terrain textures, image textures from the corresponding viewpoints were added, and texture mapping was re-performed. During the editing process, the continuity between the modified area and the surrounding model was ensured, with no obvious color difference or geometric deviation.

[0079] Secondary verification and results organization: The edited 3D model is subjected to a full-scene secondary verification. After confirming that there are no unqualified areas, the model results are systematically organized, named according to the flight strip and block specification files, stored as a standardized file package, and attached with model verification and editing instructions, recording verification results, editing methods, accuracy indicators and other information.

[0080] In this embodiment, the accuracy of the 3D model output after iterative modification reaches 96%, which fully meets the requirements for subsequent true radiometric image output.

[0081] The Photomesh software, based on the data processing system, outputs true orthophotos. The modified complete 3D model is converted into true orthophotos, serving as the core image base for wildlife identification and annotation, ensuring the accuracy of the image's geographic coordinates and visual quality. The specific steps are as follows: Image parameter settings: Set the output resolution of the true orthophoto to 0.5 meters / pixel, the output format to TIFF (including geographic coordinate information), and the projection coordinate system to WGS84 / UTM Zone 45N, which is consistent with the coordinate systems of the POS system and ArcGIS, without the need for secondary projection conversion.

[0082] True orthogonal correction: A pixel-by-pixel correction method based on a 3D model is adopted to project the texture information of the 3D model onto a planar coordinate system, eliminating image tilt and occlusion caused by the topographic undulations of the Three Rivers Source Area, and generating true orthogonal images.

[0083] Image stitching and cropping: The segmented true orthophotos are stitched together according to geographic coordinates to generate a 60km image. 2 Complete true radiographs of the work area were obtained and standardized to a size of 1km×1km to facilitate batch processing for subsequent wildlife identification.

[0084] Output: Outputs true orthophotos with complete geographic coordinates and projection information. The geographic coordinate accuracy error is ≤0.45 meters. The images can be directly imported into ArcGIS for spatial analysis and annotation.

[0085] Large wildlife identification and annotation are performed using Photomesh and ArcGIS software within the data processing system. This step is the core application of the method, relying on multi-software collaboration within the data processing system and combining precise location information from the POS system. Stereo image pairs with >65% overlap are used for stereo interpretation to achieve automatic identification, precise location, and distribution statistics for animals such as Tibetan antelopes and yaks. The specific operations are as follows: Stereoscopic interpretation and animal identification: In Photomesh software, the original high-resolution images are linked with stereo image pairs. Using a wildlife image recognition plugin, based on features such as morphology, color, and body size, groups and individual Tibetan antelopes and yaks are automatically identified. The plugin has built-in feature models of animals such as Tibetan antelopes and yaks, which can automatically mark the location and outline of animals. The recognition results can be manually reviewed. In this embodiment, the manual review correction rate is ≤4%.

[0086] Standardized sample extraction: The identified Tibetan antelope and yak image areas are automatically cropped into standardized sample images of 512×512 pixels. The sample images contain the corresponding POS geographic coordinate information and are named according to the animal species-flight strip number-image number standard to generate a standardized sample library, which can be directly used for model training and updating.

[0087] Precise location labeling: Import the true orthophoto image into the ArcGIS geographic information system, and combine it with the precise location information of the POS system to accurately label the locations of the identified Tibetan antelopes and yaks on the true orthophoto image. The labeling features are point features, and the attribute fields of the point features include: animal species, quantity, herd type (group / single), geographic coordinates (latitude and longitude / planar coordinates), and shooting time.

[0088] Distribution Map Generation and Statistical Analysis: In ArcGIS, a spatial distribution map of large herbivores is generated based on labeled point features. The distribution map is overlaid with geographical features such as water systems, topography, roads, and protected area boundaries in the Sanjiangyuan area, visually displaying the population distribution characteristics of Tibetan antelopes and yaks. At the same time, using ArcGIS's spatial analysis module, population statistical analysis is completed, including: total number, cluster size, average population density, and distribution proportion in different terrains. In this embodiment, the statistics show that there are approximately 210 Tibetan antelopes and 135 yaks in the area. Tibetan antelopes are mainly distributed in alpine meadows along the riverbanks, while yaks are mainly distributed in the valleys of river tributaries. In the population density calculation, for areas with a slope > 25°, the area correction factor is taken as 1.15.

[0089] Outputs: Outputs include wildlife identification and labeling results, a standardized sample library, spatial distribution maps, and statistical analysis reports. All outputs are in a standardized format and can be directly used for ecological research in the Three-River-Source region, wildlife conservation planning, and biodiversity assessment.

[0090] A quantitative comparison was conducted with traditional technologies across four dimensions: aerial photography efficiency, data processing accuracy, wildlife identification and positioning accuracy, and computational resource consumption. This verified the technical advantages of the present invention. The specific verification results are as follows: Aerial photography efficiency: This invention completes 60km of aerial photography in just 36 hours. 2 Compared to the 105-hour operation time of the traditional 5-lens oblique photography system combined with fragmented processing methods, the aerial photography operation in the region has increased efficiency by 65.7% and reduced the amount of aerial photography data by about 65%, significantly reducing the time and labor costs of field operations.

[0091] Data processing accuracy: The mean square error of the aerial triangulation is ≤0.14m, and the mean square error of the elevation is ≤0.19m. The accuracy of aerial triangulation is 22.5% higher than that of traditional technology, and the matching rate of corresponding points reaches 98.5%. The accuracy of the 3D model output reaches 96%, and the geographic coordinate accuracy error of the true orthographic image is ≤0.5m, which meets the requirements of 1:500 topographic mapping and high-precision ecological monitoring of the Three Rivers Source Area.

[0092] Wildlife identification and positioning accuracy: Using stereo image pairs for interpretation, the accuracy rate of identifying Tibetan antelopes and yaks reaches 96.2%, which is 30.2% higher than the accuracy rate of traditional single image interpretation and the identification efficiency is increased by 40%; the wildlife positioning error is ≤0.5 meters, which can achieve precise positioning of individual animals, far exceeding the positioning accuracy of traditional satellite remote sensing and ground inspection; the error of animal number counting is reduced to less than 5%, which is far better than the 30% error of traditional monitoring methods.

[0093] Computational resource consumption: Block computing reduces CPU usage by 25%, memory usage by 28%, and overall modeling time by 33% in 3D model building, making it suitable for distances >100km. 2 This large-scale operation area solves the problem of high computational resource consumption in traditional overall modeling methods.

[0094] In summary, this invention provides a non-invasive and refined monitoring method for large wild animals in complex terrains such as plateaus, and can be widely applied to biodiversity assessment and natural resource management.

[0095] Example 3 like Figure 8 As shown, this embodiment provides a wildlife identification and positioning system 200 based on oblique photogrammetry, including: The data acquisition module 201 is used to acquire oblique photography data for the target area. The oblique photography data includes multiple multi-view aerial images and position and attitude data corresponding to each multi-view aerial image. Each multi-view aerial image contains images acquired from multiple different directions for the same location. Image processing module 202 is used to construct a three-dimensional model of the target area based on oblique photogrammetry data, and generate a true radiometric image of the target area based on the three-dimensional model; Animal identification module 203 is used to identify wild animals based on various multi-view aerial images and obtain animal identification results; The distribution map generation module 204 is used to locate and annotate wild animals on the true radio image based on the animal identification results and the posture data of each location, so as to generate a wild animal distribution map of the target area.

[0096] Optionally, in an embodiment, the image processing module 202 includes: The preprocessing unit is used to preprocess the various multi-view aerial images to obtain preprocessed images; the preprocessing includes at least color consistency correction and geometric distortion correction. The aerial triangulation unit is used to combine the attitude data of various positions to perform aerial triangulation on the preprocessed image and generate a densified point cloud. The model generation unit is used to construct a 3D model of the target area based on encrypted point clouds and using a block-based modeling approach.

[0097] Optionally, in this embodiment, the aerial triangulation unit includes: The aerial triangulation subunit is used to perform multi-view joint adjustment on the preprocessed image. It uses a preset pyramid matching strategy to match corresponding points and combines the attitude data of each position to carry out iterative calculation of regional network adjustment to generate a densified point cloud.

[0098] Optionally, in an embodiment, the model generation unit includes: Region division sub-units are used to divide multiple sub-regions with the center point of the target region as the origin. The point cloud generation sub-unit is used to perform dense point cloud matching on each sub-region based on the encrypted point cloud, and generate high-density point clouds corresponding to each sub-region. The modeling sub-unit is used to convert the high-density point cloud of each sub-region into an irregular triangular mesh model and assign texture to it, so as to obtain the three-dimensional model of each sub-region. The model splicing sub-unit is used to splice the 3D models of various sub-regions to obtain the 3D model of the target region.

[0099] Optionally, in an embodiment, the system further includes: The stereo image pair module is used to select image pairs that meet the preset overlap conditions from various multi-view aerial images as stereo image pairs; The bounding box module is used to construct a three-dimensional bounding box for each wildlife herd area within the target area based on each stereo image pair. The first counting module is used to count the wild animals within the three-dimensional frame of each wild animal community area according to a preset animal size threshold, so as to obtain the initial count of the wild animal community area. The second counting module is used to divide the true radio image into multiple grids for each wildlife community area, count the total number of pixels belonging to wildlife in each grid, calculate the number of animals based on the total number of pixels belonging to wildlife in each grid and the pre-calibrated average number of pixels per animal, and obtain the secondary count of wildlife community areas. The quantity determination module is used to compare the difference between the initial count and the secondary count of each wildlife community area to determine the final animal population of the community area based on the degree of difference. The first information fusion module is used to integrate the final animal numbers in each wildlife population area as attribute information into the wildlife distribution map.

[0100] Optionally, in this embodiment, the quantity determination module includes: The difference calculation unit is used to calculate the difference between the initial count and the secondary count of wild animal population areas. The first determining unit is used to calculate the average of the initial count of the wild animal community area and the secondary count of the wild animal community area as the final animal count of the wild animal community area when the difference is less than or equal to the first preset threshold. The second determining unit is used to obtain frame-by-frame verification results when the difference is greater than the first preset threshold and less than or equal to the second preset threshold, so as to correct the count based on the frame-by-frame verification results and determine the final number of animals in the wildlife community area; wherein, the frame-by-frame verification results represent the animal number data obtained after performing a frame-by-frame verification of stereo image pairs in the wildlife community area; The third determining unit is used to re-count the three-dimensional bounding box of the wildlife community area when the difference is greater than the second preset threshold, so as to determine the final number of animals in the wildlife community area based on the obtained counting results.

[0101] Optionally, in an embodiment, the system further includes: The area acquisition module is used to obtain the actual geographical area of ​​each wildlife gathering area; The density calculation module is used to calculate the population density of a wildlife community based on the final number of animals and the actual geographical area of ​​the community. The second information fusion module is used to incorporate the population density of various wildlife gathering areas as attribute information into the wildlife distribution map.

[0102] In some embodiments, the wildlife identification and localization system 200 based on oblique photogrammetry of the present invention can be implemented in a combination of hardware and software. As an example, the wildlife identification and localization system 200 based on oblique photogrammetry of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the wildlife identification and localization method based on oblique photogrammetry of the present invention. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0103] The modules described in the embodiments of this invention can be implemented in software or hardware. The names of the modules are not, in some cases, limiting the scope of the module itself.

[0104] Example 4 like Figure 9 As shown, this embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the wildlife identification and positioning method based on oblique photogrammetry as described in Embodiment 1.

[0105] In other words, an electronic device according to an embodiment of the present invention may include, but is not limited to: a processor and a memory; the memory is used to store a computer program; the processor is used to execute the wildlife identification and localization method based on oblique photogrammetry shown in any embodiment of the present invention by calling the computer program.

[0106] In one alternative embodiment, an electronic device is provided. Figure 9The illustrated electronic device 300 includes a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, via a bus 302. Optionally, the electronic device 300 may further include a transceiver 304, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 304 is not limited to one type, and the structure of the electronic device 300 does not constitute a limitation on the embodiments of the present invention.

[0107] Processor 301 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 301 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0108] Bus 302 may include a path for transmitting information between the aforementioned components. Bus 302 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 302 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 9 The bus 302 is represented by only one thick line, but this does not mean that there is only one bus or one type of bus.

[0109] The memory 303 may be a ROM (Read Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0110] The memory 303 is used to store application code (computer program) for executing the present invention, and its execution is controlled by the processor 301. The processor 301 is used to execute the application code stored in the memory 303 to implement the content shown in the foregoing method embodiments.

[0111] Among them, electronic devices can also be terminal devices, which can be any device that can install applications, including at least one of smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, smart TVs, and smart in-vehicle devices.

[0112] It should be noted that, Figure 9 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0113] Example 5 This embodiment provides a non-transitory computer-readable storage medium that stores computer instructions for causing a computer to execute the wildlife identification and localization method based on oblique photogrammetry of Embodiment 1.

[0114] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0115] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the aforementioned method for wildlife identification and localization based on oblique photogrammetry.

[0116] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0117] It should be understood that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0118] The computer-readable storage medium provided in this invention can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EEPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0119] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the method shown in the above embodiments.

[0120] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0121] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0122] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this invention can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, this invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0123] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for recognizing and locating wild animals based on oblique photogrammetry, characterized in that, include: Oblique photography data for a target area is acquired. The oblique photography data includes multiple multi-view aerial images and position and attitude data corresponding to each multi-view aerial image. Each multi-view aerial image contains images acquired from multiple different directions for the same location. A three-dimensional model of the target area is constructed based on the oblique photogrammetry data, and a true radiometric image of the target area is generated based on the three-dimensional model. Wildlife identification is performed based on aerial images from various multi-view perspectives to obtain animal identification results; Based on the animal identification results and the posture data of each location, wildlife location annotations are performed on the true radio image to generate a wildlife distribution map of the target area.

2. The method for wildlife identification and localization based on oblique photogrammetry according to claim 1, characterized in that, The step of constructing a three-dimensional model of the target area based on the oblique photogrammetry data includes: The aerial images from various multi-view perspectives are preprocessed to obtain preprocessed images; wherein the preprocessing includes at least color consistency correction and geometric distortion correction. By combining the attitude data from various positions, aerial triangulation is performed on the preprocessed image to generate a densified point cloud; Based on the encrypted point cloud, a three-dimensional model of the target region is constructed using a block-based modeling approach.

3. The tilt photography-based wild animal recognition and positioning method according to claim 2, characterized in that, The step of combining the pose data from various positions to perform aerial triangulation on the preprocessed image to generate a densified point cloud includes: Multi-view joint adjustment is performed on the preprocessed images. A preset pyramid matching strategy is used to match corresponding points. Iterative calculations of regional network adjustment are carried out in combination with the pose data of each location to generate a densified point cloud. 4.The wild animal recognizing and positioning method based on oblique photogrammetry according to claim 2, characterized in that, The construction of a 3D model of the target region based on the encrypted point cloud, using a block-based modeling approach, includes: Divide the target region into multiple sub-regions with the center point of the target region as the origin; Based on the encrypted point cloud, dense point cloud matching is performed on each sub-region to generate a high-density point cloud corresponding to each sub-region. The high-density point cloud of each sub-region is converted into an irregular triangular mesh model and textured to obtain the 3D model of each sub-region. The 3D models of each sub-region are stitched together to obtain the 3D model of the target region.

5. The tilt photography-based wild animal recognition and positioning method according to claim 1, characterized in that, Also includes: Image pairs that meet the preset overlap conditions are selected from various multi-view aerial images as stereo image pairs; Based on each stereo image pair, construct a three-dimensional stereo outline of each wildlife community within the target area; For each wildlife community area, the number of wild animals within the three-dimensional frame of the wildlife community area is counted according to a preset animal size threshold to obtain the initial count of the wildlife community area. For each wildlife community area, multiple grids are divided on the true radio image. The total number of pixels belonging to wildlife in each grid is counted. The number of animals is calculated based on the total number of pixels belonging to wildlife in each grid and the pre-defined average number of pixels per animal, thus obtaining the secondary count of the wildlife community area. For each wildlife colony area, the difference between the initial count and the secondary count of the wildlife colony area is compared to determine the final number of animals in the wildlife colony area based on the degree of difference. The final animal population in each wildlife colony area is incorporated as attribute information into the wildlife distribution map.

6. The tilt photography-based wild animal recognition and positioning method according to claim 5, characterized in that, For each wildlife colony, the difference between the initial count and the secondary count of the wildlife colony is compared to determine the final animal population of the wildlife colony based on the degree of difference, including: Calculate the difference between the initial count and the secondary count of the wildlife community area; When the difference is less than or equal to the first preset threshold, the average of the initial count of the wild animal community area and the secondary count of the wild animal community area is calculated as the final animal count of the wild animal community area. When the difference is greater than the first preset threshold and less than or equal to the second preset threshold, a frame-by-frame verification result is obtained to correct the count based on the frame-by-frame verification result and determine the final number of animals in the wildlife community area; wherein, the frame-by-frame verification result represents the animal number data obtained after performing a frame-by-frame verification of stereo image pairs in the wildlife community area; When the difference is greater than the second preset threshold, the three-dimensional frame selection and counting of the wildlife community area is performed again to determine the final number of animals in the wildlife community area based on the obtained counting results.

7. The tilt photography-based wild animal recognition and positioning method according to claim 5, characterized in that, Also includes: Obtain the actual geographical area of ​​each wildlife colony; Calculate the population density of the wildlife community based on the final number of animals and the actual geographical area of ​​the community. The population density of each wildlife community area is incorporated as attribute information into the wildlife distribution map.

8. A tilt photography based wildlife identification and location system, characterized in that, include: The data acquisition module is used to acquire oblique photography data for the target area. The oblique photography data includes multiple multi-view aerial images and position and attitude data corresponding to each multi-view aerial image. Each multi-view aerial image contains images acquired from multiple different directions for the same location. An image processing module is used to construct a three-dimensional model of the target area based on the oblique photogrammetry data, and to generate a true radiometric image of the target area based on the three-dimensional model. The animal identification module is used to identify wild animals based on aerial images from various multi-view perspectives and obtain animal identification results. The distribution map generation module is used to perform wildlife location annotation on the true radio image based on the animal identification results and the posture data of each location, so as to generate a wildlife distribution map of the target area.

9. An electronic device, comprising: The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the wildlife identification and localization method based on oblique photogrammetry as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to execute the wildlife identification and localization method based on oblique photogrammetry as described in any one of claims 1 to 7.

Images