CN114066999B - Target positioning system and method based on three-dimensional modeling - Google Patents

Abstract

The invention discloses a three-dimensional modeling-based target positioning system and a three-dimensional modeling-based target positioning method, which belong to the technical field of emergency rescue, wherein an image recognition module in the system recognizes a target on each image in original image data through an artificial intelligent algorithm; the target analysis module in the system analyzes and judges whether the positions of the targets on each image on the ground are the same, the targets on each image are reckoned after the targets are de-duplicated, and then the reckoned targets are positioned on a ground three-dimensional model and displayed; the target is directly identified from the original image data through an artificial intelligence algorithm, and the target position is calculated and then subjected to de-duplication processing, so that the target data volume is effectively reduced, the same target is prevented from being counted for multiple times, the accuracy of target counting is improved, and the counted target position is displayed on a three-dimensional model generated by the same image data, so that the target is positioned more intuitively, and the necessary support is more favorably provided for rapid auxiliary decision making of emergency management.

Description

Target positioning system and method based on three-dimensional modeling

Technical Field

The present invention relates to the technical field of emergency rescue, and more specifically, mainly to a target positioning system and method based on three-dimensional modeling.

Background Art

In recent years, natural disasters have occurred frequently in various parts of my country, which has put forward a great challenge to the timeliness and efficiency of emergency rescue. After a natural disaster occurs, how to quickly obtain the terrain of the disaster area and the accurate location of the victims is the key to making emergency rescue decisions. Since most of the current mainstream target positioning methods are not developed for emergency rescue scenarios, the positioning time is long, the data redundancy is large, and the positioning data cannot be displayed intuitively after acquisition, which makes it unsuitable for use in emergency rescue scenarios. In view of this, it is necessary to study and improve the methods of rapid positioning of victims and display used in emergency rescue scenarios that rely on computers.

Summary of the invention

One of the purposes of the present invention is to address the above-mentioned shortcomings and to provide a target positioning system and method based on three-dimensional modeling, in the hope of solving the technical problems of similar target positioning methods in the prior art, such as long positioning time, large data redundancy, inability to intuitively display positioning data, and unsuitability for use in emergency rescue scenarios.

In order to solve the above technical problems, the present invention adopts the following technical solutions:

The first aspect of the present invention provides a target positioning system based on three-dimensional modeling, and the system includes: an image recognition module, which is used to identify the target on each image in the original image data through an artificial intelligence algorithm; a target analysis module, which is used to analyze and determine whether the positions of the targets on each image are the same on the ground. If they are the same, it is considered that the current two or more targets are the same target; the analysis and judgment of whether the positions of the targets on each image are the same on the ground is to compare whether the distance between the positions of the targets on the ground exceeds a specified threshold. If it does not exceed, it is considered that the positions of the targets on the ground are the same, otherwise it is considered that the current two or more targets are not the same target; the target analysis module is also used to deduplicate the targets on each image and re-count them, and then locate the re-counted targets on the ground three-dimensional model for display; the ground three-dimensional model is generated by the original image data; the deduplication is to count two or more targets judged to be the same target as one target.

As a preferred embodiment, a further technical solution is: the position of the target on the ground is: assuming that the target on each image is point P, the ray direction v passing through the camera center point and point P is obtained by the following formula:

v＝R^'K^(-1)P (1)

In formula (1), R is the image posture rotation matrix, and K is the camera intrinsic parameter matrix; thus, the parameter equation of any point on the ray starting from the camera center and passing through point P is obtained:

P＝c+μv (2)

In formula (2), c is the vector of the image position, and μ is the distance between the point on the ray and the center of the camera. Through the ray and space triangle intersection algorithm, the point I where the ray intersects with the triangle of the ground three-dimensional model is determined and obtained, thereby obtaining the coordinate value of the current target position on the ground.

A further technical solution is: the target is a person in each image.

A further technical solution is that the original image data is a plurality of images taken by a drone and compressed.

A further technical solution is: the system is used to locate people in disaster areas during emergency management and provide support for rescue decisions.

A second aspect of the present invention provides a target positioning method based on three-dimensional modeling, wherein the method comprises the following steps:

Identify the target on each image in the original image data through artificial intelligence algorithms;

Analyze and judge whether the positions of the targets on each image are the same on the ground. If they are the same, it is considered that the two or more targets are the same target. The analysis and judgment of whether the positions of the targets on each image are the same on the ground is to compare whether the distance between the positions of the targets on the ground exceeds a specified threshold. If not, it is considered that the positions of the targets on the ground are the same. Otherwise, it is considered that the two or more targets are not the same target.

The targets on each image are deduplicated and re-counted, and then the re-counted targets are positioned and displayed on a three-dimensional ground model; the three-dimensional ground model is generated by the original image data; the deduplication is to count two or more targets judged to be the same target as one target.

As a preferred embodiment, a further technical solution is: the position of the target on the ground is assuming that the target on each image is point P, and the ray direction v passing through the camera center point and point P is obtained by the following formula:

v＝R^'K^(-1)P (1)

In formula (1), R is the image posture rotation matrix, and K is the camera intrinsic parameter matrix; thus, the parameter equation of any point on the ray starting from the camera center and passing through point P is obtained:

P＝c+μv (2)

In formula (2), c is the vector of the image position, μ is the distance between the point on the ray and the camera center;

Through the ray and space triangle intersection algorithm, the point I where the ray intersects with the triangle of the ground three-dimensional model is determined and obtained, so as to obtain the coordinate value of the current target position on the ground.

A further technical solution is: the target is a person in each image; the original image data is a plurality of images taken by a drone and compressed.

A further technical solution is: the system is used to locate people in disaster areas during emergency management and provide support for rescue decisions.

A third aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, and when a computer executes the instructions, the computer executes the above method.

Compared with the prior art, one of the beneficial effects of the present invention is that the target is directly identified from the original image data through an artificial intelligence algorithm, and deduplication processing is performed after the target position is calculated, thereby effectively reducing the amount of target data and avoiding the same target being counted multiple times, thereby improving the accuracy of target statistics, and by displaying the counted target position on the three-dimensional model generated by the same image data, the positioning of the target is more intuitive, which is more conducive to providing necessary support for rapid auxiliary decision-making in emergency management.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic block diagram of a system for illustrating an embodiment of the present invention.

FIG. 2 is a schematic block diagram of the target analysis module structure for illustrating an embodiment of the present invention.

FIG. 3 is a flowchart for illustrating a method according to an embodiment of the present invention.

FIG. 4 is a flow chart of a method for generating a ground three-dimensional model according to an embodiment of the present invention.

DETAILED DESCRIPTION

The present invention will be further described below in conjunction with the accompanying drawings.

The present invention identifies the target from the original image data, and then realizes the target positioning through calculation, and directly displays the positioned target on the three-dimensional model, and the three-dimensional model is a three-dimensional model that is quickly generated under the technical requirements of emergency management. On the basis of the foregoing summary, one embodiment of the present invention is a target positioning system based on three-dimensional modeling, which is mainly used for locating people in disaster areas in emergency management and providing support for rescue decisions. Specifically, as shown in FIG1, according to the functional division of the modules, the system mainly includes two parts, namely an image recognition module and a target analysis module, and data can be transmitted between the image recognition module and the target analysis module.

Specifically, the above-mentioned image recognition module identifies the target on each image in the original image data through an artificial intelligence algorithm (AI). The target recognition can refer to the similar AI video recognition technology in the prior art. Generally speaking, in the application of this system, the target on the image identified by AI is a person, more specifically, a disaster victim. And the above-mentioned original image data generally refers to multiple images taken by drones and compressed.

After the target data in the image is obtained in the above manner, it is transmitted to the target analysis module, which analyzes and determines whether the positions of the targets in each image are the same on the ground. If they are the same, the two or more targets are considered to be the same target. The above-mentioned ground position refers to the ground coordinates, and the ground coordinates belong to the same coordinate system as the above-mentioned three-dimensional model.

In the target analysis module, an operation is performed to analyze and determine whether the positions of the targets on each image are the same on the ground. Specifically, it refers to comparing whether the distance between the positions of the targets on the ground exceeds a specified threshold. If not, the positions of the targets on the ground are considered to be the same. Otherwise, the two or more targets are considered to be different targets and need to be counted separately.

As mentioned above, in order to prevent the same target in multiple images from being counted multiple times, resulting in excessive target data, the target analysis module needs to deduplicate the targets on each image and then re-count them. The aforementioned deduplication means that when two or more targets are judged to be the same target, the current two or more targets are counted as one target.

Specifically, for two targets T1 and T2 on two images taken at similar times, assuming that their positions on the ground are calculated to be L1 and L2 respectively, and assuming that the distance between L1 and L2 is less than a specified threshold, then T1 and T2 are considered to be the same target. Otherwise, T1 and T2 are counted as two targets, and the above deduplication process is not performed on T1 and T2. Since the target of the present invention is people in the disaster area, the displacement distance of the same person in two images taken at similar times is usually relatively small, so the above assumption can be considered reasonable.

The target analysis module positions the re-counted targets on the ground three-dimensional model. The ground three-dimensional model is generated by the original image data, that is, the original image data of the AI-recognized target.

As mentioned above, the ground coordinates of the target positioning are in the same coordinate system as the three-dimensional model, so after the target positioning, it can be directly displayed with the three-dimensional model as the background, that is, the target positioning is displayed on the ground three-dimensional model.

In the above embodiment, the target analysis module first needs to know the position of each target on the ground before performing the above deduplication judgment. This embodiment provides a preferred method, where the target analysis module can obtain the position of the target on the ground, that is, the ground coordinates, from the original image.

Specifically, let the target on each image be point P, and the ray direction v passing through the camera center point and point P is obtained by the following formula:

v＝R^'K^(-1)P (1)

In formula (1), R is the image posture rotation matrix, and K is the camera intrinsic parameter matrix. The aforementioned R and K can be calculated by the original image data through the aerotriangulation optimization algorithm. In this embodiment, the two are equivalent to known quantities; and then the parameter equation of any point on the ray starting from the camera center and passing through point P is obtained:

P＝c+μv (2)

In formula (2), c is the vector of the image position, and μ is the distance between the point on the ray and the camera center. As mentioned above, point P in formula (2) refers to any point on the ray that starts from the camera center and passes through point P, which naturally includes point P itself.

Through the ray and space triangle intersection algorithm, the point I where the ray intersects with the triangle of the ground three-dimensional model is determined and obtained, thereby obtaining the coordinate value of the current target on the ground. Specifically, since the three-dimensional model represents the ground terrain, the intersection of the ray and the triangle in the three-dimensional model also means the intersection with the ground, and then when the target is a point on the ray, the coordinate value of the target on the ground can be obtained.

The above ray and space triangle intersection algorithm can be implemented by referring to the following process:

Assume that a point is located on the triangle (V0, V1, V2), where V0, V1, V2 are the coordinates of the three vertices of the triangle. Then this point can be represented, for example, in the following way:

T(u,v)＝(1-u-v)×V0+u×V1+v×V2

Here u+v≤1,u≥0,v≥0;

For rays, we generally use the following equation to represent it:

R(t)=O+t×D (R is the image posture rotation matrix, O is the starting point of the ray, and D is the direction of the ray)

So, since they have to have an intersection, we can directly use the following method to get it:

O+t×D＝(1-u-v)×V0+u×V1+v×V2

The above equations form a system of three linear equations, which can be solved for the three parameters t, u and v. If 0<u<1, 0<v<1, and u+v≤1, then the point is inside the triangle, which means that the ray intersects the triangle. Otherwise, the two do not intersect.

In the above three-variable linear equation system, u and v are two parameters of the point on the plane where the triangle is located expressed by the coordinates of the three vertices of the triangle. u represents the normalized component of the point on the line segments V0 and V1, and v represents the normalized component of the point on the line segments V0 and V2. That is, if the intersection point is inside the triangle, then 0<u<1, 0<v<1, and u+v≤1. If the above conditions are not met, the intersection point is outside the triangle. t is a parameter of a point on the ray, and t represents the directed distance of the point from the starting point of the ray. t>0 means that the point is in the positive direction of the ray. t<0 means that it is in the negative direction of the ray.

To ensure that the present invention can be fully disclosed to those skilled in the art, with reference to FIG2 , another embodiment of the present invention is a hardware structure of a target analysis module for analyzing whether targets are identical and locating targets in the above embodiment. In this embodiment, it includes a processor, a memory, and a data interface. The target on each image in the original image data is received through the data interface and stored in the memory. The memory is also used to store necessary computer instructions. The processor reads the target on each image in the memory and executes corresponding computer instructions, thereby realizing the functions of the target analysis module in the above embodiment.

In a specific implementation, the processor may include one or more central processing units (CPUs), where a processor may refer to one or more devices, circuits, and/or processing cores for processing data (eg, computer program instructions).

The memory may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory may exist independently and be connected to the processor through a bus. The memory may also be integrated with the processor.

The data interface may use any type of computer hardware data interface commonly used in the art, so that the image data can be received and written to a storage medium readable by a processor in the intelligent modeling unit.

Referring to FIG. 3 , based on the above system, another embodiment of the present invention is a target positioning method based on three-dimensional modeling, the method comprising the following steps:

Step S101: Identify the target on each image in the original image data by artificial intelligence algorithm. The method is applied in emergency management scenarios, so the aforementioned target generally refers to the person on each image in the original image data, especially the disaster victims; and the original image data mainly comes from multiple images taken by drones and compressed.

Step S102, respectively analyzing and calculating the position of each target on the ground;

Step S103: determine whether the targets in each image are the same. If they are the same, it is considered that the current two or more targets are the same target.

Specifically, we analyze and determine whether the positions of the targets on the ground in each image are the same, and compare whether the distances between the positions of the targets on the ground exceed a specified threshold. If not, we consider that the positions of the targets on the ground are the same; otherwise, we consider that the two or more targets are not the same target and need to be counted separately.

Step S104: remove duplicates from the objects in each image and re-count them, then locate and display the re-counted objects on the ground three-dimensional model. The above-mentioned removal of duplicates is to count two or more objects that are determined to be the same object as one object.

Specifically, the above-mentioned ground three-dimensional model is generated by the original image data.

Preferably, the position of the target on the ground is obtained by the following steps:

Step S1041, assuming that the target on each image is point P, the ray direction v passing through the camera center point and point P is obtained by the following formula:

v＝R^'K^(-1)P (1)

In formula (1), R is the image posture rotation matrix, and K is the camera intrinsic parameter matrix;

Step S1042: Using the ray direction v obtained above, construct a parametric equation for any point on the ray starting from the camera center and passing through point P:

P＝c+μv (2)

In formula (2), c is the vector of the image position, and μ is the distance between the ray and the camera center. As mentioned above, point P in formula (2) refers to any point on the ray starting from the camera center and passing through point P, which naturally includes point P itself.

Step S1043: By using the ray and space triangle intersection algorithm, determine and obtain the point I where the ray intersects with the triangle of the ground three-dimensional model, thereby obtaining the coordinate value of the current target position on the ground. The ray and space triangle intersection algorithm in this step can be implemented with reference to the method of the above embodiment.

Based on the general form of computer software products, another embodiment of the present invention provides a computer-readable storage medium, in which instructions are stored. When a computer executes the instructions, the computer executes the target positioning method based on three-dimensional modeling in the above embodiment.

Among them, the computer-readable storage medium, for example, can be but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, RAM, ROM, an erasable programmable read only memory (EPROM), a register, a hard disk, an optical fiber, a CD-ROM, an optical storage device, a magnetic storage device, or any suitable combination of the above, or any other form of computer-readable storage medium known in the art. An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an application-specific integrated circuit (ASIC). In an embodiment of the present application, the computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

More preferably, the above-mentioned ground three-dimensional model is generated by the original image data, and the ground coordinates of the target are in the same coordinate system as the ground three-dimensional model. Therefore, after the target is located, it is displayed with the three-dimensional model as the background. The two can be directly superimposed to achieve the positioning and display of people in the disaster area. The target display is more intuitive and reliable, and is more conducive to providing necessary support for rapid decision-making assistance for emergency management.

In the above-mentioned multiple embodiments of the present invention, the position of targets such as people in disaster areas on the ground can be accurately located through the position and posture information of the image and the three-dimensional model of the surface, effectively reducing the possibility of the same target being counted multiple times in overlapping images and increasing the accuracy of target statistics.

Referring to FIG4 , the above-mentioned three-dimensional ground model generated based on the original image data can be quickly generated by the following method:

Step S001, obtain compressed original image data, and construct a scene graph through the compressed image data; when constructing the scene graph, based on the spatial position (longitude and latitude) of the image when it is acquired and according to the spatial neighbor principle, select a certain number (for example, 50) of images closest to each image as its neighboring images, so as to quickly construct the scene graph.

Step S002, using SIFT (Scale-invariant feature transform) or similar algorithms to extract feature points on each image to complete image matching, the feature points are the front-most feature points after the feature points on each image are arranged from large to small according to scale. Optionally, only a certain number (e.g., 2000-4000) of feature points with the largest scale are retained instead of using all feature points for feature point matching, which can greatly reduce the matching time.

Step S003, determine the position and posture information of each image when it was taken through an aerial triangulation optimization algorithm; the aerial triangulation optimization algorithm is a commonly used algorithm in image processing, which finds the optimal camera position and posture information and the three-dimensional coordinates of the feature points based on the matched feature points, thereby minimizing the sum of squares of the reprojection errors of the three-dimensional points.

Step S004, using the stereo pair matching algorithm to generate a dense point cloud from the position and posture information of each image when it was taken; as shown in FIG4 , the dense point cloud generation is to downsample each image by a preset image downsampling coefficient, and specify multiple point cloud sampling density levels, and only generate a full pixel depth map for the image at a high density level; the aforementioned image downsampling coefficient is generally 2 by default. The aforementioned stereo pair matching algorithm is also a commonly used algorithm in image processing, which can determine the same-name pixels of each pixel and the adjacent image according to the positional relationship between each image and the adjacent image, and then use the forward intersection algorithm to determine the position of the corresponding pixel, thereby obtaining a dense point cloud.

Optionally, the above-mentioned point cloud sampling density levels include high density level, medium density level and low density level; except for the high density level, the full pixel depth map is generated for the image, and the other levels are all generated by interval pixels, that is, at the medium density level, the depth map is generated for every other pixel in the horizontal and vertical directions of the image; at the low density level, the depth map is generated for every two pixels in the horizontal and vertical directions of the image. By specifying the downsampling coefficient and point cloud sampling density, the number and generation time of dense point clouds can be greatly reduced, which also saves time for the subsequent 3D model generation.

Step S005: Generate a triangulated mesh model based on the generated dense point cloud; then perform texture mapping between the triangulated mesh model and the image data to generate a three-dimensional model close to the real-time image of the modeling area. The triangulated mesh model is generated based on the generated dense point cloud by using a Delaunay tetrahedron partitioning algorithm and a graph cut method.

In this step, it is optional to first generate a Delaunay tetrahedron spatial partition based on the generated dense point cloud, construct a global optimization graph, the nodes of the graph are constructed by tetrahedrons in the spatial partition, and the edges of the graph are triangular faces of adjacent tetrahedrons, then determine the triangular faces in the Delaunay tetrahedron partition where the lines from each point to the camera it sees intersect, and add a weight value of 1 to the corresponding edge in the global optimization graph to obtain a global optimization graph with visible line constraints, and finally use the maximum flow minimum cut algorithm to segment the global optimization graph, determine the internal and external relationship between each tetrahedron and the model surface, and extract the shared triangular faces of adjacent tetrahedrons located inside and outside the surface to form the final triangulated network model. Then select the visible camera closest to each triangular face as its associated camera, and then divide the spatially connected triangular faces of the same associated camera into the same group, obtain the image blocks taken on the associated camera, and finally combine all the image blocks into a texture map according to the packing algorithm to realize the texture mapping of the triangular network, and obtain the above-mentioned three-dimensional model that is the same or close to the real-time image of the modeling area.

In addition to the above, it should be noted that "one embodiment", "another embodiment", "embodiment", etc. mentioned in this specification refer to the specific features, structures or characteristics described in conjunction with the embodiment included in at least one embodiment generally described in this application. The same expression appearing in multiple places in the specification does not necessarily refer to the same embodiment. Further, when describing a specific feature, structure or characteristic in conjunction with any embodiment, it is claimed that the realization of such feature, structure or characteristic in conjunction with other embodiments also falls within the scope of the present invention.

Although the present invention is described herein with reference to a number of illustrative embodiments of the present invention, it will be appreciated that those skilled in the art may devise many other modifications and implementations that fall within the scope and spirit of the principles disclosed herein. More specifically, within the scope of the present disclosure, drawings, and claims, a variety of variations and improvements may be made to the components and/or layout of the subject combination layout. In addition to variations and improvements made to the components and/or layout, other uses will also be apparent to those skilled in the art.

Claims (8)

1. A three-dimensional modeling-based object localization system, said system comprising:

the image recognition module is used for recognizing the target on each image in the original image data through an artificial intelligence algorithm;

the target analysis module is used for respectively analyzing and judging whether the positions of the targets on each image are the same or not, and if the positions of the targets on the ground are the same, the current targets are considered to be the same targets;

the analysis judges whether the positions of the targets on each image on the ground are the same or not, if not, the positions of the targets on the ground are the same, otherwise, the current targets are not the same;

the target analysis module is also used for carrying out re-statistics on targets on each image after the targets are de-duplicated, and then positioning the re-counted targets on a ground three-dimensional model for display;

the ground three-dimensional model is generated through the original image data; the de-duplication is to count a plurality of targets which are judged to be the same target as one target;

the position of the target on the ground is as follows: let the target on each image be the P point, obtain the ray direction v passing through the center point of the camera and the P point by the following formula:

v=R^' K^(-1) P （1）

in the formula (1), R is an image attitude rotation matrix, and K is a camera internal parameter matrix; and then obtaining a parameter equation of any point on the ray which starts from the center of the camera and passes through the point P:

P=c+μv （2）

in the formula (2), c is a vector of image positions, and μ is a distance value between a point on the ray and a center of the camera;

And judging and obtaining a point I at which the ray intersects with the triangle of the ground three-dimensional model through a ray and space triangle intersection algorithm, thereby obtaining the coordinate value of the current target at the ground position.

2. The three-dimensional modeling-based object localization system of claim 1, wherein: the target is a person on each image.

3. The three-dimensional modeling-based object localization system of claim 1, wherein: the original image data are a plurality of images shot and compressed by the unmanned aerial vehicle.

4. The three-dimensional modeling-based object localization system of claim 1, wherein: the system is used for positioning disaster area personnel in emergency management and providing support for rescue decisions.

5. The target positioning method based on three-dimensional modeling is characterized by comprising the following steps of:

Identifying a target on each image in the original image data through an artificial intelligence algorithm;

respectively analyzing and judging whether the positions of the targets on each image on the ground are the same, if so, considering the current targets as the same target; the analysis judges whether the positions of the targets on each image are the same or not, if not, the positions of the targets on the ground are the same, otherwise, the current targets are not the same;

Re-counting the targets on each image after the targets are de-duplicated, and then positioning the re-counted targets on a ground three-dimensional model for display; the ground three-dimensional model is generated through the original image data; the de-duplication is to count a plurality of targets which are judged to be the same target as one target;

The position of the target on the ground is set as a P point, and the ray direction v passing through the center point and the P point of the camera is obtained by the following formula:

v=R^' K^(-1) P （1）

in the formula (1), R is an image attitude rotation matrix, and K is a camera internal parameter matrix; and then obtaining a parameter equation of any point on the ray which starts from the center of the camera and passes through the point P:

P=c+μv （2）

in the formula (2), c is a vector of image positions, and μ is a distance value between a point on the ray and a center of the camera;

And judging and obtaining a point I at which the ray intersects with the triangle of the ground three-dimensional model through a ray and space triangle intersection algorithm, thereby obtaining the coordinate value of the current target at the ground position.

6. The three-dimensional modeling-based object localization method of claim 5, wherein: the target is a person on each image; the original image data are a plurality of images shot and compressed by the unmanned aerial vehicle.

7. The three-dimensional modeling-based object localization method of claim 5, wherein: the method is used for positioning disaster area personnel in emergency management and providing support for rescue decisions.

8. A computer-readable storage medium, characterized by: the computer-readable storage medium having instructions stored therein, which when executed by a computer, cause the computer to perform the method of any of claims 5 to 7.