Video twinning and target detection collaborative LLM intelligent data system

Abstract

The invention discloses a video twinning and target detection collaborative LLM intelligent data system, relates to the field of intelligent security and protection, solves the problem of poor detection effect of an existing security and protection target detection system, and comprises a model creation module which creates a target detection area into a three-dimensional space model and divides the three-dimensional space model into a plurality of image rendering areas, the three-dimensional rendering module is used for matching an optimal rendering camera for an image rendering area according to a three-dimensional space coordinate system, and creating a three-dimensional rendering model of a target area by performing image real-time rendering on the image rendering area to obtain model creation data, and the target detection module is used for performing image real-time monitoring on the three-dimensional rendering model of the target area to obtain model creation data. The image rendering module divides the plurality of image rendering areas into abnormal behavior areas and non-abnormal behavior areas according to monitoring results to obtain target behavior detection data, and the intelligent early warning module carries out intelligent early warning on the target detection areas according to the target behavior detection data.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent security and relates to video twin technology, specifically an LLM intelligent data system that collaborates with video twin and target detection. Background Art

[0002] The existing target detection systems used in the field of intelligent security have the following specific defects when performing security anomaly detection: 1. The existing target detection system can only detect security anomalies through two-dimensional surveillance images. Since two-dimensional surveillance images lack depth information, it is difficult to accurately judge the target distance, height and volume, resulting in a lack of accuracy in the detection results; 2. The existing target detection system is unable to create corresponding video image detection models for specific monitoring areas, resulting in a lack of specificity in the detection process.

[0003] To this end, we propose an LLM intelligent data system that collaborates with video twins and object detection. Summary of the invention

[0004] In view of the shortcomings of the prior art, the purpose of the present invention is to provide an LLM intelligent data system that collaborates with video twins and target detection. The present invention aims to improve the accuracy and diversity of security anomaly detection.

[0005] In order to achieve the above objectives, the present invention adopts the following technical solution: an LLM intelligent data system for coordinated video twinning and target detection, comprising: Model creation module: Create a three-dimensional space model of the target detection area and divide it into multiple image rendering areas, create a three-dimensional space coordinate system in the three-dimensional space model, match the best rendering camera for each image rendering area according to the three-dimensional space coordinate system, use the best rendering camera to perform real-time image rendering on each image rendering area to create a three-dimensional rendering model of the target area, and obtain model creation data; Target detection module: performs real-time image monitoring of the three-dimensional rendering model of the target area according to the model creation data, divides multiple image rendering areas into abnormal behavior areas and non-abnormal behavior areas according to the monitoring results, and obtains target behavior detection data; Smart security field: Provide intelligent early warning for target detection areas based on target behavior detection data.

[0006] Furthermore, the model creation data is obtained as follows: In the target detection area, a laser radar is used to perform a three-dimensional scan of the target detection area, and a three-dimensional space model of the target detection area is created according to the scanning result; Create a three-dimensional coordinate system in the three-dimensional space model to obtain the three-dimensional space coordinate system; The three-dimensional space model is divided into a plurality of image rendering areas, and a sample image rendering area is selected from the divided plurality of image rendering areas; Selecting a surveillance camera for the sample image rendering area to obtain the best rendering camera corresponding to the sample image rendering area; Obtain the best rendering camera corresponding to each image rendering area respectively; In the three-dimensional space model, the monitoring image acquired by each optimal rendering camera is rendered in real time to the corresponding image rendering area through the image rendering algorithm to obtain a three-dimensional rendering model of the target area; A plurality of image rendering areas, a target area three-dimensional rendering model, and a three-dimensional space coordinate system are defined as model creation data.

[0007] Furthermore, a three-dimensional space coordinate system is created as follows: In a three-dimensional space model, arbitrarily select a vertex of the bottom surface of the model as the coordinate origin, and mark the bottom surface of the model as the first model feature plane. In the first model feature plane, draw a straight line through the coordinate origin to obtain the coordinate x-axis, draw a straight line perpendicular to the coordinate x-axis through the coordinate origin to obtain the coordinate y-axis, draw a plane perpendicular to the first model feature plane through the coordinate x-axis to obtain the second model feature plane, and in the second model feature plane, draw a straight line perpendicular to the coordinate x-axis to obtain the coordinate z-axis. The plane composed of the coordinate x-axis, the coordinate z-axis and the coordinate origin is marked as the three-dimensional space coordinate system.

[0008] Furthermore, the best rendering camera corresponding to the sample image rendering area is obtained, as follows: Marking the surveillance cameras installed in the three-dimensional control model as valid surveillance cameras and invalid surveillance cameras; Randomly select a valid surveillance camera as the sample surveillance camera, and obtain the coordinate distance value between the sample filling pixel point and the sample surveillance camera: The coordinate distance value between each edge filling pixel point and the sample monitoring camera is obtained respectively to obtain a plurality of coordinate distance values, and the average of the obtained plurality of coordinate distance values ​​is calculated to obtain the effective monitoring distance value between the sample monitoring camera and the sample image rendering area; The effective monitoring distance value between each effective monitoring camera and the sample image rendering area is obtained to obtain multiple effective monitoring distance values, and the multiple effective monitoring distance values ​​are compared, and the effective monitoring camera with the smallest effective monitoring distance is marked as the best rendering camera corresponding to the sample image rendering area.

[0009] Furthermore, valid surveillance cameras and invalid surveillance cameras are marked as follows: The monitoring cameras installed in the three-dimensional control model are acquired to obtain multiple model monitoring cameras. Each model monitoring camera is used to acquire monitoring images of the sample image rendering area to obtain multiple sample area monitoring images. If the sample area monitoring image can capture a complete image of the sample image rendering area, the model monitoring camera corresponding to the sample area monitoring image is marked as a valid monitoring camera. If the sample area monitoring image cannot capture a complete image of the sample image rendering area, the model monitoring camera corresponding to the sample area monitoring image is marked as an invalid monitoring camera.

[0010] Furthermore, the coordinate distance value between the sample filling pixel point and the sample monitoring camera is obtained, as follows: In the three-dimensional space coordinate system, the coordinate points corresponding to the sample monitoring camera are marked as camera feature points; Filling pixels on the edge of the sample image rendering area to obtain a plurality of edge filling pixels, and selecting a sample filling pixel from the plurality of edge filling pixels; Obtain the three-dimensional coordinates of the sample filling pixel point in the three-dimensional space coordinate system to obtain the first feature three-dimensional coordinates, obtain the three-dimensional coordinates of the camera feature point in the three-dimensional space coordinate system to obtain the second feature three-dimensional coordinates; The first characteristic three-dimensional coordinate and the second characteristic three-dimensional coordinate are calculated to obtain a coordinate distance value between the sample filling pixel point and the sample monitoring camera; The coordinate distance value is calculated, and the specific formula is as follows: ; Among them, Zbj is the coordinate distance value, (x1, y1, z1) is the first feature three-dimensional coordinate, and (x2, y2, z2) is the second feature three-dimensional coordinate.

[0011] Furthermore, the target behavior detection data is obtained as follows: Acquire model creation data, acquire a three-dimensional rendering model of a target area, a three-dimensional space coordinate system, and a plurality of image rendering areas according to the model creation data, and render the model on the acquired plurality of images; Acquire a real-time three-dimensional image corresponding to each image rendering area through a three-dimensional rendering model of the target area; Create a large model for target behavior detection, use the large model to detect the real-time three-dimensional image corresponding to each image rendering area, mark the image rendering area where the target detection behavior image exists as the abnormal behavior area, and obtain the coordinates of the target behavior position in the three-dimensional space coordinate system to obtain the abnormal detection behavior coordinates, mark the image rendering area where the target detection behavior image does not exist as the non-abnormal behavior area, and obtain the target behavior detection data.

[0012] Furthermore, the target behavior detection large model is created as follows: Acquire multiple historical three-dimensional images corresponding to the three-dimensional rendering models of the target area, select a sample historical three-dimensional image from the multiple historical three-dimensional images acquired, and classify the sample historical three-dimensional image into image types; Marking target detection behavior for each historical three-dimensional image respectively to obtain historical three-dimensional image marking data; The historical 3D image labeled data is divided into a model 3D image training set and a model 3D image test set according to the image training and testing ratio; Create an image recognition model through an existing artificial intelligence platform, and train the image recognition model using a model 3D image training set until the image recognition model is trained once for each model 3D image in the model 3D image training set; The image recognition model is tested using the model three-dimensional image test set, and the recognition accuracy is obtained. When the recognition accuracy is greater than or equal to the target recognition accuracy, the image recognition model training is completed and the target behavior detection model is obtained. When the recognition accuracy is less than the target recognition accuracy, the image recognition model is continued to be trained using the model three-dimensional image training set until the recognition accuracy is greater than or equal to the target recognition accuracy.

[0013] Furthermore, the sample historical three-dimensional images are divided into image types as follows: Use image recognition algorithms to mark pedestrians in sample historical three-dimensional images and obtain multiple pedestrian models; In the three-dimensional space coordinate system, the z-axis coordinate value of each pedestrian model is obtained to obtain the height value of the pedestrian model, the area covered by each pedestrian model in the first model feature plane is obtained to obtain the area value of the pedestrian model, and the ratio of the height value of each pedestrian model to the area value of the pedestrian model is calculated to obtain multiple model height-to-area ratios; The preset high-occupancy area ratio interval when the human body is in a non-lying state is obtained through the LLM intelligent model; If any model has a high area ratio that is not in the preset high area ratio range, the sample historical three-dimensional image is marked as a target detection behavior image; if any model has a high area ratio that is in the preset high area ratio range, the sample historical three-dimensional image is marked as a non-target detection behavior image.

[0014] Furthermore, the intelligent security early warning for the target detection area is specifically as follows: Acquire target behavior detection data, and acquire abnormal behavior areas and non-abnormal behavior areas respectively according to the target behavior detection data; If there is an abnormal behavior area in the target detection area, an abnormal behavior warning is issued for the target detection area, and the abnormal detection behavior coordinates are automatically output; If there is no abnormal behavior area in the target detection area, no abnormal behavior warning will be issued for the target detection area.

[0015] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. The present invention creates a target detection area as a three-dimensional space model and divides it into multiple image rendering areas, creates a three-dimensional space coordinate system in the three-dimensional space model, matches the best rendering camera for each image rendering area according to the three-dimensional space coordinate system, uses the best rendering camera to perform real-time image rendering on each image rendering area to create a three-dimensional rendering model of the target area, and performs security abnormal behavior detection through the three-dimensional rendering model of the target area, which can effectively improve the accuracy of the detection result; 2. The present invention creates a large target behavior detection model based on the three-dimensional rendering model of the target area to perform real-time image monitoring, and divides multiple image rendering areas into abnormal behavior areas and non-abnormal behavior areas according to the monitoring results, which can effectively ensure the environmental targeting of the detection process. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0017] Figure 1 is the overall system block diagram of the present invention; Figure 2 It is a schematic diagram of the three-dimensional space coordinate system in the present invention; Figure 3 Schematic diagram of camera feature points in the present invention. DETAILED DESCRIPTION

[0018] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] See also Figure 1 , the present invention provides a technical solution: a LLM intelligent data system for collaborative video twin and target detection, comprising a model creation module, a target detection module, an intelligent early warning module and a server, wherein the model creation module, the target detection module and the intelligent early warning module are respectively connected to the server, and the server controls the model creation module, the target detection module and the intelligent early warning module respectively; The model creation module creates the target detection area as a three-dimensional space model and divides it into multiple image rendering areas, creates a three-dimensional space coordinate system in the three-dimensional space model, matches the best rendering camera for each image rendering area according to the three-dimensional space coordinate system, uses the best rendering camera to perform real-time image rendering on each image rendering area to create a three-dimensional rendering model of the target area, and obtains model creation data; The details are as follows: In the target detection area, a laser radar is used to perform a three-dimensional scan of the target detection area, and a three-dimensional space model of the target detection area is created according to the scanning result; See also Figure 2 In the three-dimensional space model, a vertex of the bottom surface of the model is randomly selected as the origin of the coordinate, and the bottom surface of the model is marked as the first model characteristic plane. In the first model characteristic plane, a straight line is randomly drawn through the origin of the coordinate to obtain the coordinate x-axis. A straight line perpendicular to the coordinate x-axis is drawn through the origin of the coordinate to obtain the coordinate y-axis. A plane perpendicular to the first model characteristic plane is drawn through the coordinate x-axis to obtain the second model characteristic plane. In the second model characteristic plane, a straight line perpendicular to the coordinate x-axis is drawn to obtain the coordinate z-axis. The plane composed of the coordinate x-axis, the coordinate z-axis and the origin of the coordinate is marked as the three-dimensional space coordinate system. It should be noted here that: In the present application, the first model feature plane involved here is the ground surface of the target detection area.

[0020] The three-dimensional space model is divided into a plurality of image rendering areas, and a sample image rendering area is selected from the divided plurality of image rendering areas; Selecting a surveillance camera for the sample image rendering area to obtain the best rendering camera corresponding to the sample image rendering area; The details are as follows: The monitoring cameras installed in the three-dimensional control model are acquired to obtain a plurality of model monitoring cameras, and each model monitoring camera is used to acquire monitoring images of the sample image rendering area to obtain a plurality of sample area monitoring images. If the sample area monitoring image can take a complete image of the sample image rendering area, the model monitoring camera corresponding to the sample area monitoring image is marked as a valid monitoring camera; if the sample area monitoring image cannot take a complete image of the sample image rendering area, the model monitoring camera corresponding to the sample area monitoring image is marked as an invalid monitoring camera; It should be noted here that: In the present application, the multiple model surveillance cameras involved here are all cameras of the same model and the same parameters.

[0021] See also Figure 3 , arbitrarily select a valid surveillance camera as a sample surveillance camera, and mark the coordinate points corresponding to the sample surveillance camera as camera feature points in the three-dimensional space coordinate system; Filling pixels on the edge of the sample image rendering area to obtain a plurality of edge filling pixels, and selecting a sample filling pixel from the plurality of edge filling pixels; Obtain the three-dimensional coordinates of the sample filling pixel point in the three-dimensional space coordinate system to obtain the first feature three-dimensional coordinates, obtain the three-dimensional coordinates of the camera feature point in the three-dimensional space coordinate system to obtain the second feature three-dimensional coordinates; The first characteristic three-dimensional coordinate and the second characteristic three-dimensional coordinate are calculated to obtain a coordinate distance value between the sample filling pixel point and the sample monitoring camera; The coordinate distance value is calculated, and the specific formula is as follows: ; Among them, Zbj is the coordinate distance value, (x1, y1, z1) is the first feature three-dimensional coordinate, (x2, y2, z2) is the second feature three-dimensional coordinate; In the specific implementation, there are the following experimental data: If the first feature 3D coordinates are (3, 66, 23), and if the second feature 3D coordinates are (87, 56, 73), then the coordinate distance can be calculated to be 99.27m; If the first feature three-dimensional coordinates are (32, 63, 33), and if the second feature three-dimensional coordinates are (27, 36, 33), then the coordinate distance can be calculated to be 27.466m; If the three-dimensional coordinates of the first feature are (35, 67, 23) and the three-dimensional coordinates of the second feature are (87, 86, 83), then the coordinate distance can be calculated to be 81.64m.

[0022] It should be noted here that: Here, the above distance calculation formula is used to calculate the precise three-dimensional coordinates, so as to obtain a more accurate coordinate distance value between the sample filling pixel point and the sample monitoring camera; The distance value between the rendering target area and the surveillance camera is represented according to the coordinate distance value, and the camera with the closest distance is selected and marked as the best rendering camera. The image obtained by the best rendering camera is used to render the model. Marking the camera with the closest distance as the best rendering camera can ensure that the image used is more matched with the object in the scene, thereby improving the visual quality of the rendering effect, reducing unnecessary distortion, and at the same time reducing unnecessary computing resource consumption and optimizing system performance.

[0023] Repeat the process of obtaining the coordinate distance value between the sample filling pixel point and the sample monitoring camera, respectively obtain the coordinate distance value between each edge filling pixel point and the sample monitoring camera, obtain multiple coordinate distance values, calculate the average of the obtained multiple coordinate distance values, and obtain the effective monitoring distance value between the sample monitoring camera and the sample image rendering area; Repeat the process of obtaining the effective monitoring distance value between the sample monitoring camera and the sample image rendering area, respectively obtain the effective monitoring distance value between each effective monitoring camera and the sample image rendering area, obtain multiple effective monitoring distance values, and compare the values ​​of the obtained multiple effective monitoring distance values, and mark the effective monitoring camera with the smallest effective monitoring distance as the best rendering camera corresponding to the sample image rendering area; It should be noted here that: In the present application, if there are multiple effective monitoring cameras with the smallest effective monitoring distance, one of them is arbitrarily selected as the best rendering camera.

[0024] Repeat the process of obtaining the best rendering camera corresponding to the sample image rendering area, and obtain the best rendering camera corresponding to each image rendering area respectively; In the three-dimensional space model, the monitoring image acquired by each optimal rendering camera is rendered in real time to the corresponding image rendering area through the image rendering algorithm to obtain a three-dimensional rendering model of the target area; defining a plurality of image rendering areas, a target area three-dimensional rendering model, and a three-dimensional space coordinate system as model creation data; The model creation module acquires the model creation data and transmits it to the target detection module and the intelligent warning module; The target detection module performs real-time image monitoring of the three-dimensional rendering model of the target area according to the model creation data, and divides multiple image rendering areas into abnormal behavior areas and non-abnormal behavior areas according to the monitoring results to obtain target behavior detection data; The details are as follows: Acquire model creation data, acquire a three-dimensional rendering model of a target area, a three-dimensional space coordinate system, and a plurality of image rendering areas according to the model creation data, and render the model on the acquired plurality of images; Acquire a real-time three-dimensional image corresponding to each image rendering area through the target area three-dimensional rendering model to obtain multiple real-time three-dimensional images; Acquire multiple historical three-dimensional images corresponding to the three-dimensional rendering models of the target area, and create a large model for target behavior detection based on the multiple historical three-dimensional images; The details are as follows: A sample historical three-dimensional image is selected from the multiple historical three-dimensional images obtained, and a pedestrian in the sample historical three-dimensional image is marked as a model using an image recognition algorithm to obtain multiple pedestrian models; In the three-dimensional space coordinate system, the z-axis coordinate value of each pedestrian model is obtained to obtain the height value of the pedestrian model, the area covered by each pedestrian model in the first model feature plane is obtained to obtain the area value of the pedestrian model, and the ratio of the height value of each pedestrian model to the area value of the pedestrian model is calculated to obtain multiple model height-to-area ratios; The preset high-occupancy area ratio interval when the human body is in a non-lying state is obtained through the LLM intelligent model; It should be noted here that: When a normal pedestrian is in a non-lying state, the reasonable range of the pedestrian model height-to-area ratio (i.e., the ratio of height to floor area) is usually between 0.1 and 10. The specific analysis process is as follows: When standing or walking normally, the height of a pedestrian (from the top of the head to the ground) is usually between 1.5 meters and 2 meters; The footprint (i.e. the projected area of ​​a pedestrian on a horizontal plane) is affected by body posture and is approximately 0.5 to 1 square meter when standing.

[0025] Therefore, the high footprint ratio is approximately: Height (1.7m) / footprint (0.3m²) ≈ 5.6 Taking into account actual posture changes (such as arm swinging, body leaning forward, etc.), the ratio will be appropriately reduced, so 0.1 to 10 is taken as the typical range.

[0026] If any model has a high area ratio that is not in the preset high area ratio interval, the sample historical three-dimensional image is marked as a target detection behavior image; if any model has a high area ratio that is in the preset high area ratio interval, the sample historical three-dimensional image is marked as a non-target detection behavior image; Repeat the process of marking the target detection behavior of the sample historical three-dimensional image, mark the target detection behavior of each historical three-dimensional image respectively, and obtain the historical three-dimensional image marking data; The historical 3D image labeled data is divided into a model 3D image training set and a model 3D image test set according to the image training and testing ratio; It should be noted here that: In the present application, the image training test ratio is specifically set to 7:3, that is, the ratio of the number of model 3D images in the model 3D image training set and the number of model 3D image test sets is 7:3; Create an image recognition model through an existing artificial intelligence platform, and train the image recognition model using a model 3D image training set until the image recognition model is trained once for each model 3D image in the model 3D image training set; The image recognition model is tested using the model three-dimensional image test set, and the recognition accuracy is obtained. When the recognition accuracy is greater than or equal to the target recognition accuracy, the image recognition model training is completed and the target behavior detection model is obtained. When the recognition accuracy is less than the target recognition accuracy, the image recognition model is continued to be trained using the model three-dimensional image training set until the recognition accuracy is greater than or equal to the target recognition accuracy.

[0027] It should be noted here that: The target recognition accuracy involved here is specifically set to 95% in this application.

[0028] Use the target behavior detection large model to detect the real-time three-dimensional image corresponding to each image rendering area, mark the image rendering area where the target detection behavior image exists as the abnormal behavior area, and obtain the coordinates of the target behavior position in the three-dimensional space coordinate system to obtain the abnormal detection behavior coordinates, mark the image rendering area where the target detection behavior image does not exist as the non-abnormal behavior area, and obtain the target behavior detection data; The target behavior involved here is specifically a pedestrian lying in the target detection area. In this application, a pedestrian lying in the target detection area corresponds to specific security events including but not limited to medical emergencies, traffic accidents, and pedestrian mental health problems. The various security events listed above are all abnormal security events.

[0029] The target detection module acquires the target behavior detection data and transmits it to the intelligent warning module; The intelligent warning module provides intelligent warning for the target detection area based on the target behavior detection data; The details are as follows: Acquire target behavior detection data, and acquire abnormal behavior areas and non-abnormal behavior areas respectively according to the target behavior detection data; If there is an abnormal behavior area in the target detection area, an abnormal behavior warning is issued for the target detection area, and the abnormal detection behavior coordinates are automatically output; If there is no abnormal behavior area in the target detection area, no abnormal behavior warning will be issued for the target detection area.

[0030] In this application, if corresponding calculation formulas appear, the above calculation formulas are all dimensionless and take their numerical calculations. The weight coefficients, proportional coefficients and other coefficients in the formulas are set to a result value obtained by quantifying each parameter. The size of the weight coefficient and the proportional coefficient can be determined as long as it does not affect the proportional relationship between the parameter and the result value.

[0031] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well.

Claims

1. An LLM intelligent data system for video twinning and target detection, characterized in that: include: Model creation module: Create a three-dimensional space model of the target detection area and divide it into multiple image rendering areas, create a three-dimensional space coordinate system in the three-dimensional space model, match the best rendering camera for each image rendering area according to the three-dimensional space coordinate system, use the best rendering camera to perform real-time image rendering on each image rendering area to create a three-dimensional rendering model of the target area, and obtain model creation data; Target detection module: performs real-time image monitoring of the three-dimensional rendering model of the target area according to the model creation data, divides multiple image rendering areas into abnormal behavior areas and non-abnormal behavior areas according to the monitoring results, and obtains target behavior detection data; Intelligent warning module: Provide intelligent warning for the target detection area based on the target behavior detection data.

2. According to claim 1, a video twin and target detection collaborative LLM intelligent data system is characterized in that: Get the model creation data as follows: In the target detection area, a three-dimensional scan is performed on the target detection area, and a three-dimensional space model is created for the target detection area according to the scanning result; Create a three-dimensional coordinate system in the three-dimensional space model to obtain the three-dimensional space coordinate system; The three-dimensional space model is divided into a plurality of image rendering areas, and a sample image rendering area is selected from the divided plurality of image rendering areas; Select a surveillance camera for each image rendering area to obtain the best rendering camera corresponding to each image rendering area; In the three-dimensional space model, the monitoring image acquired by each optimal rendering camera is used to perform real-time image rendering on the corresponding image rendering area to obtain a three-dimensional rendering model of the target area; A plurality of image rendering areas, a target area three-dimensional rendering model, and a three-dimensional space coordinate system are defined as model creation data.

3. According to claim 2, a video twin and target detection collaborative LLM intelligent data system is characterized in that: Create a three-dimensional space coordinate system as follows: In a three-dimensional space model, arbitrarily select a vertex of the bottom surface of the model as the coordinate origin, and mark the bottom surface of the model as the first model feature plane. In the first model feature plane, draw a straight line through the coordinate origin to obtain the coordinate x-axis, draw a straight line perpendicular to the coordinate x-axis through the coordinate origin to obtain the coordinate y-axis, draw a plane perpendicular to the first model feature plane through the coordinate x-axis to obtain the second model feature plane, and in the second model feature plane, draw a straight line perpendicular to the coordinate x-axis to obtain the coordinate z-axis. The plane composed of the coordinate x-axis, the coordinate z-axis and the coordinate origin is marked as the three-dimensional space coordinate system.

4. According to claim 2, the LLM intelligent data system for video twin and target detection collaboration is characterized in that: The best rendering camera corresponding to the sample image rendering area is obtained as follows: Marking the surveillance cameras installed in the three-dimensional control model as valid surveillance cameras and invalid surveillance cameras; Randomly select a valid surveillance camera as the sample surveillance camera, and obtain the coordinate distance value between the sample filling pixel point and the sample surveillance camera: The coordinate distance value between each edge filling pixel point and the sample monitoring camera is obtained to obtain multiple coordinate distance values, and the average of the obtained multiple coordinate distance values ​​is calculated to obtain the effective monitoring distance value between the sample monitoring camera and the sample image rendering area; The effective monitoring distance value between each effective monitoring camera and the sample image rendering area is obtained to obtain multiple effective monitoring distance values, and the multiple effective monitoring distance values ​​are compared, and the effective monitoring camera with the smallest effective monitoring distance is marked as the best rendering camera corresponding to the sample image rendering area.

5. According to claim 2, the LLM intelligent data system for video twin and target detection collaboration is characterized in that: Valid surveillance cameras and invalid surveillance cameras are marked as follows: The monitoring cameras installed in the three-dimensional control model are acquired to obtain multiple model monitoring cameras. Each model monitoring camera is used to acquire monitoring images of the sample image rendering area to obtain multiple sample area monitoring images. If the sample area monitoring image can capture a complete image of the sample image rendering area, the model monitoring camera corresponding to the sample area monitoring image is marked as a valid monitoring camera. If the sample area monitoring image cannot capture a complete image of the sample image rendering area, the model monitoring camera corresponding to the sample area monitoring image is marked as an invalid monitoring camera.

6. According to claim 4, a video twin and target detection collaborative LLM intelligent data system is characterized in that: The coordinate distance value between the sample filling pixel point and the sample monitoring camera is obtained as follows: In the three-dimensional space coordinate system, the coordinate points corresponding to the sample monitoring camera are marked as camera feature points; Filling pixels at the edge of the sample image rendering area, and selecting a sample filling pixel from a plurality of edge filling pixels; Obtain the three-dimensional coordinates of the sample filling pixel point in the three-dimensional space coordinate system to obtain the first feature three-dimensional coordinates, obtain the three-dimensional coordinates of the camera feature point in the three-dimensional space coordinate system to obtain the second feature three-dimensional coordinates; The first characteristic three-dimensional coordinate and the second characteristic three-dimensional coordinate are calculated to obtain a coordinate distance value between the sample filling pixel point and the sample monitoring camera; The coordinate distance value is calculated, and the specific formula is as follows: ； Among them, Zbj is the coordinate distance value, (x1, y1, z1) is the first feature three-dimensional coordinate, and (x2, y2, z2) is the second feature three-dimensional coordinate.

7. According to claim 1, the LLM intelligent data system for video twin and target detection collaboration is characterized in that: The target behavior detection data is obtained as follows: Acquire model creation data, acquire a three-dimensional rendering model of a target area, a three-dimensional space coordinate system, and a plurality of image rendering areas according to the model creation data, and render the model on the acquired plurality of images; Acquire a real-time three-dimensional image corresponding to each image rendering area through a three-dimensional rendering model of the target area; Create a target behavior detection large model, use the target behavior detection large model to detect the real-time three-dimensional image corresponding to each image rendering area, mark the image rendering area where the target detection behavior image exists as the abnormal behavior area, and obtain the coordinates of the target behavior position in the three-dimensional space coordinate system to obtain the abnormal detection behavior coordinates, and mark the image rendering area where the target detection behavior image does not exist as the non-abnormal behavior area; Get target behavior detection data.

8. The LLM intelligent data system for video twin and target detection collaboration according to claim 6 is characterized in that: Create a large model for target behavior detection, as follows: Acquire multiple historical three-dimensional images corresponding to the three-dimensional rendering models of the target area, select a sample historical three-dimensional image from the multiple historical three-dimensional images acquired, and classify the sample historical three-dimensional image into image types; Marking target detection behavior for each historical three-dimensional image respectively to obtain historical three-dimensional image marking data; The historical 3D image labeled data is divided into a model 3D image training set and a model 3D image test set according to the image training and testing ratio; Create an image recognition model through an existing artificial intelligence platform and train the image recognition model using a model 3D image training set; The image recognition model is tested using the model three-dimensional image test set, and the recognition accuracy is obtained. When the recognition accuracy is greater than or equal to the target recognition accuracy, the image recognition model training is completed and the target behavior detection model is obtained. When the recognition accuracy is less than the target recognition accuracy, the image recognition model is continued to be trained using the model three-dimensional image training set until the recognition accuracy is greater than or equal to the target recognition accuracy.

9. The LLM intelligent data system for video twin and target detection collaboration according to claim 8 is characterized in that: The sample historical three-dimensional images are divided into image types as follows: Model marking is performed on pedestrians in sample historical three-dimensional images to obtain multiple pedestrian models; In the three-dimensional space coordinate system, the z-axis coordinate value of each pedestrian model is obtained to obtain the height value of the pedestrian model, the area covered by each pedestrian model in the first model feature plane is obtained to obtain the area value of the pedestrian model, and the ratio of the height value of each pedestrian model to the area value of the pedestrian model is calculated to obtain multiple model height-to-area ratios; The preset high-occupancy area ratio interval when the human body is in a non-lying state is obtained through the LLM intelligent model; If any model has a high area ratio that is not in the preset high area ratio range, the sample historical three-dimensional image is marked as a target detection behavior image; if any model has a high area ratio that is in the preset high area ratio range, the sample historical three-dimensional image is marked as a non-target detection behavior image.

10. The LLM intelligent data system for video twin and target detection collaboration according to claim 1 is characterized in that: Provide intelligent security warning for the target detection area, as follows: Acquire target behavior detection data, and acquire abnormal behavior areas and non-abnormal behavior areas respectively according to the target behavior detection data; If there is an abnormal behavior area in the target detection area, an abnormal behavior warning is issued for the target detection area, and the abnormal detection behavior coordinates are automatically output; If there is no abnormal behavior area in the target detection area, no abnormal behavior warning will be issued for the target detection area.

Images