A method for identifying a target in a wide area

Abstract

The present application relates to the field of image target recognition, and particularly relates to a wide range target identity recognition method, comprising: establishing a virtual-real double space unified ground plane polar coordinate system with the ground plane projection of the camera optical center point as the origin and the ground plane projection direction of the field of view optical axis of the camera as the polar axis; calculating the polar angle and polar radius of each target in the polar coordinate system respectively; screening out the target matching the visual image information and the geographic information, and associating the screened target with the target identity information in the target management system; the present application has the advantages that: the ground plane polar coordinate system is established with the ground plane projection of the camera optical center point as the origin and the ground plane projection direction of the field of view optical axis as the polar axis, the two-stage matching judgment of the polar angle and the polar radius is realized, the accurate association of the visual target and the geographic information target is realized, the false detection rate and the false recognition rate of the identity recognition are significantly reduced, and the present application is used for the complex monitoring scene of the long distance, the multiple targets and the similar appearance.

Description

Technical Field

[0001] This invention relates to the field of image target recognition, and more particularly to a method for target identification over a wide area. Background Technology

[0002] In wide-area surveillance scenarios, such as open-pit mines, airports, smart ports, and large construction sites, real-time monitoring and identification of multiple targets over long distances and large areas is often required. However, in these scenarios, targets (such as vehicles, ships, and aircraft) are far from the camera, have small image sizes and blurred features, and similar targets often look alike, making identification using only computer vision technology extremely difficult and costly. Therefore, a multimodal fusion system is urgently needed for joint identification to achieve target identification and to correlate and query multiple known information related to the identity (such as vehicle number, license plate number, driver, load status, etc.).

[0003] In target recognition and tracking systems based on fixed-position cameras, traditional methods typically have the following limitations: (1) Lack of multimodal fusion capability: Existing methods mostly focus on single computational models (such as pure visual recognition or pure positioning systems), failing to effectively integrate the advantages of visual recognition and geolocation systems, making it difficult to achieve accurate target identity association. Visual systems can provide rich image information, but cannot directly obtain the target's identity identifier; geolocation systems (such as GNSS, AIS, ADS-B) can provide the target's location and identity information, but cannot be intuitively displayed in video footage. The lack of effective means to integrate the two limits the overall performance of the system.

[0004] (2) Limitations in long-distance visual recognition capabilities: At the edge of the field of view or at medium to long distances, the target image size is small, features are blurred, and the target features are highly similar with few unique features. Relying solely on visual AI algorithms (such as face recognition, license plate recognition, and vehicle re-identification) for identity recognition significantly increases the false detection and false recognition rates, making it difficult to meet the requirements of high-reliability applications. Especially in harsh environments (such as dust, low light, rain, fog, and vibration), the stability of visual recognition further decreases.

[0005] (3) Multi-target and multi-modal recognition is costly; In scenarios requiring continuous monitoring and identification of numerous targets over a large area, deploying a separate high-precision vision and positioning fusion system for each target or sub-region would significantly increase hardware costs, computing power consumption, and system complexity, making it difficult to achieve cost-effective large-scale deployment and application. For example, equipping each vehicle with a separate high-precision positioning and vision recognition terminal would be too costly and inconvenient for centralized management.

[0006] (4) Insufficient real-time performance and system stability; Existing fusion methods often employ complex spherical geometric calculations or high-precision geodetic models, resulting in high computational loads and response delays, making it difficult to meet the needs of real-time monitoring of multiple targets. Furthermore, they lack robust handling mechanisms for practical engineering problems such as positioning signal interruptions, visual recognition failures, and computational anomalies, making it difficult to guarantee system stability. Summary of the Invention

[0007] The purpose of this invention is to provide a target identification method for a wide area, which solves the problems of low visual recognition accuracy, poor multimodal fusion capability, high computational cost and insufficient system stability in existing technologies in long-distance, multi-target monitoring scenarios. This invention establishes a unified virtual and real dual-space ground plane polar coordinate system and combines a two-stage matching strategy of polar angle and polar radius to achieve accurate and rapid association between visual targets and geographic information targets, thereby improving the accuracy, real-time performance and robustness of target identification, and reducing system deployment and computing power costs.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A method for target identification over a wide area includes: S1. Simultaneously acquire the target image coordinates collected by the visual recognition system and the target geographic coordinates provided by the target management system; S2. Establish a ground plane polar coordinate system that unifies the virtual and real spaces, with the projection of the camera's optical center point onto the ground plane as the origin and the projection direction of the camera's field of view optical axis onto the ground plane as the polar axis. S3. Calculate the polar angle and polar radius of each target in the polar coordinate system; S4. Match and judge the polar angle and polar radius, filter out targets whose visual image information matches the geographic information, and associate the filtered targets with the target identity information in the target management system.

[0009] In S3, the polar angles include: The first polar angle is calculated based on the target's pixel x-coordinate in the panoramic video image, the image's horizontal center line, and the horizontal field of view. The second polar angle is calculated based on the camera's latitude and longitude, the target's latitude and longitude, and the camera's geographic orientation angle.

[0010] The expression for the first polar angle is: ; in: The x-coordinate of the target pixel, in pixels; The image width is in pixels. The horizontal field of view is expressed in radians.

[0011] The second polar angle includes: The latitude and longitude coordinates are rounded to a preset precision. By approximating the Earth's surface as a local plane, the northward and eastward components of the target relative to the camera are calculated. The geographic azimuth is calculated based on the north and east components, and the camera orientation angle is subtracted from the geographic azimuth to obtain the horizontal deflection angle relative to the camera's optical axis, which is used as the second polar angle.

[0012] The expression for the northbound component is: ; The expression for the eastward component is: ; The expression for the second polar angle is: ; This is the arctangent function for the four quadrants, with the output unit being degrees (after passing through). (After conversion) , The target latitude and longitude are rounded down, in degrees; , The latitude and longitude of the camera are rounded down to the nearest degree. This is the average latitude in radians, expressed in degrees. This is the camera's geographic orientation angle, in degrees. This refers to the horizontal field of view, in degrees. This is a function that standardizes angles to a specified range.

[0013] In S3, the polar radius includes: The first polar radius is calculated based on the target's pixel coordinates in the panoramic video image, image height, vertical field of view, camera pitch angle, and installation height. The second polar radius is calculated based on the Euclidean distance between the target and the camera, which is the difference in planar coordinates.

[0014] In S3: the expression for the first polar radius is: ; in: The installation height of the camera is specified in meters. The vertical field of view is expressed in radians. Image height, in pixels; The pixel distance is derived from the pixel's ordinate, in pixels. The expression for the second polar radius is: ; in: The eastward component is in meters; The northward component is represented by meters.

[0015] In S4: The matching judgment between polar angle and polar radius includes: Calculate the difference between the first polar angle and the second polar angle. ,if If the match is successful, then proceed to polar radius matching; otherwise, the matching is considered to have failed. Calculate the difference between the first polar diameter and the second polar diameter. ,if If the match is successful, the target match is considered successful; otherwise, the match is considered unsuccessful. in: The preset angle deviation threshold, in degrees; This is the preset distance deviation threshold, in meters.

[0016] It also includes an exception handling mechanism: When the target positioning signal is interrupted, based on the velocity vector and direction vector before the target was lost, according to the time interval... The estimated location is expressed as follows: ; in: The estimated target location is shown in meters. The target position at the last moment before the signal was lost, in meters; The velocity vector of the target, in meters per second; The time interval is estimated in seconds; The identity display is maintained through the above dead reckoning; When visual recognition fails but the positioning data shows that the target is within the effective range, a speculative marker box is displayed at the estimated position on the screen. Set a watchdog timer for each match calculation; if the calculation times out, force termination and discard abnormal data.

[0017] It also includes augmented reality demonstration steps: Bind the successfully matched target identity information to the visual tracking ID; The identification mark and operation status information are overlaid and displayed at the corresponding target location in the video frame.

[0018] Compared with the prior art, the beneficial effects of the present invention are: 1. By synchronously acquiring the target image coordinates of the visual recognition system and the target geographic coordinates of the target management system, a ground plane polar coordinate system is established with the projection of the camera's optical center point onto the ground plane as the origin and the projection direction of the field of view optical axis onto the ground plane as the polar axis. This coordinate system unifies the visual space and geographic space under the same geometric framework. Through two-stage matching judgment of polar angle and polar radius, the accurate association between visual targets and geographic information targets is achieved. Compared with existing single vision or single positioning schemes, this invention significantly reduces the false detection rate and false recognition rate of identity recognition, and is especially suitable for complex monitoring scenarios with long distances, multiple targets, and similar appearances. 2. A local planar approximation is adopted to replace the complex spherical geodetic model, converting latitude and longitude coordinates into plane rectangular coordinates, which greatly reduces the amount of calculation. At the same time, a two-stage matching strategy is adopted, using polar angle and polar radius matching to filter out most invalid targets. This architecture significantly improves the computational efficiency while ensuring matching accuracy, meeting the response time requirements of real-time monitoring of multiple targets. 3. Integrates multiple anomaly handling and self-recovery mechanisms: Positioning signal interruption handling: When the target GNSS signal is briefly lost, the system performs short-term dead reckoning based on the motion vector before the loss, maintaining the identity display. It automatically corrects itself after the signal is restored, preventing identity flickering or loss; Visual recognition failure handling: When AI target detection fails due to dust, low light, etc., if the positioning data shows the target is within the effective range, the system displays a semi-transparent "predicted marker box" at the estimated position on the screen, preventing complete information loss; Calculation process protection: A watchdog timer is set for each matching calculation. If the timeout occurs, the system forcibly terminates and discards abnormal data, ensuring the overall system throughput capacity. These mechanisms enable the invention to maintain stable operation even in harsh environments such as dust, vibration, temperature differences, and light variations, demonstrating strong engineering practicality. 4. There is no need to deploy high-precision recognition equipment separately for each target or sub-area. Only the data fusion engine of the visual recognition system and the target management system needs to be deployed in the monitoring center to achieve unified recognition and management of multiple targets in a wide area. It has low hardware cost, low computing power consumption, and high system integration. It is suitable for application scenarios that require large-scale deployment, such as open mines, airports, ports, large construction sites, and security perimeters. 5. Bind the successfully matched target identity information with the visual tracking ID, and overlay key information such as device number, vehicle speed, and load status on the video screen to achieve augmented reality effect. Monitoring personnel can intuitively grasp the status of on-site equipment without having to consult multiple systems, which significantly improves dispatching efficiency and operational safety. 6. All successfully matched events (including target ID, identity information, matching timestamp, geographic coordinates, video slice index, etc.) are structured and stored in the database. This record is not only used for real-time monitoring, but also provides a complete data foundation for production efficiency analysis, backtracking of work safety incidents, and verification of scheduling instruction execution. It has good traceability and scalability. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of a target identity recognition application scenario covering a wide area.

[0020] Figure 2 It is a panoramic video AI target image.

[0021] Figure 3 This is a side view of the camera's position.

[0022] Figure 4 This is the architecture diagram of the target geospatial relationship determination system.

[0023] Figure 5 This is a flowchart for determining the geospatial relationship of the target. Detailed Implementation

[0024] The present invention will now be described in detail with reference to the accompanying drawings, but it should be noted that the implementation of the present invention is not limited to the following embodiments.

[0025] The following embodiments are implemented based on the technical solution of the present invention, providing detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments. Unless otherwise specified, the methods used in the following embodiments are conventional methods.

[0026] Example 1:

[0027] A method for target identification over a wide area includes: S1. Simultaneously acquire the target image coordinates collected by the visual recognition system and the target geographic coordinates provided by the target management system; See Figure 1 Taking an open-pit mine monitoring scenario as an example, a high-definition panoramic camera is deployed at the mine entrance and exit. The camera parameters are as follows: image resolution 1920×1080 pixels, horizontal field of view... (Approximately 1.0472 radians), vertical field of view (Approximately 0.6109 radians), camera installation height meters, pitch angle (Approximately 0.0873 radians), Geographical Orientation Angle (With true north as 0°, increase clockwise).

[0028] The visual recognition system includes: a high-definition panoramic camera, a video capture card, a deep learning-based target detection algorithm module (such as YOLO or Faster R-CNN), a target tracking module, and an image coordinate output interface. The system performs real-time AI target detection on the video stream, identifies target vehicles in the image, and outputs their image coordinates.

[0029] See Figure 2 The panoramic video AI target image shows that the target vehicle E has pixel coordinates (X,Y)=(1240,540) pixels in the image.

[0030] Meanwhile, the target management system, such as the open-pit mine vehicle dispatching system, obtains the vehicle's geographical coordinates via GNSS, which are then rounded down to: pointLat1 = 40.1235°, where pointLat represents the target dimension; pointLon1 = 116.5678°, where pointLon represents the target longitude.

[0031] The geographical coordinates of the camera installation site are: cameraLat1 = 40.1200°, where cameraLat represents the camera dimension. cameraLon1 = 116.5650°, where cameraLon represents the longitude of the camera.

[0032] The vehicle's identification information includes: equipment number "MT-1023", speed 25km / h, and load status "heavy load".

[0033] S2. Establish a ground plane polar coordinate system that unifies the virtual and real spaces, with the projection of the camera's optical center point onto the ground plane as the origin and the projection direction of the camera's field of view optical axis onto the ground plane as the polar axis. See Figure 2 and Figure 3 A polar coordinate system is established with the origin O as the vertical projection point P of the camera's optical center on the ground plane and the polar axis direction as the projection direction of the camera's field of view optical axis on the ground plane.

[0034] The specific setup process is as follows: See Figure 3 The camera was mounted at a height The optical center of the pole at the top of the meter. The vertical projection point on the ground plane is P. The angle between the camera's optical axis and the horizontal line is the pitch angle. The projection of this optical axis onto the ground plane is the polar axis direction, with the geographic orientation angle as the reference point. (Southeast direction) is the positive polar axis. A polar coordinate system is established with P as the origin, the polar axis as the reference direction, and counterclockwise as the positive direction. In this coordinate system, the position of any target can be uniquely represented by its polar angle (the angle of deflection relative to the polar axis) and polar radius (the straight-line distance to the origin P).

[0035] This polar coordinate system is used to describe both visually identified targets (virtual space) and targets in geographic information (real space), achieving the unification of virtual and real spaces.

[0036] S3. Calculate the polar angle and polar radius of each target in the polar coordinate system; The calculation of the polar angle S31 includes: S311. Calculate the first polar angle based on the target's pixel x-coordinate in the panoramic video image, the image's horizontal center line, and the horizontal field of view. ; The expression for the first polar angle is: ; in: The x-coordinate of the target pixel. Pixel; The image width in pixels. Pixel; For horizontal field of view, radian.

[0037] radian.

[0038] S312. Calculate the second polar angle based on the camera's latitude and longitude, the target's latitude and longitude, and the camera's geographic orientation angle.

[0039] The second polar angle includes: (1) Round the latitude and longitude coordinates to a preset precision (retain 4 decimal places). ; ; ; .

[0040] (2) Approximate the Earth's surface as a local plane and calculate the northward and eastward components of the target relative to the camera; The expression for the northbound component is: ; The expression for the eastward component is: ; It represents the number of meters corresponding to each degree of latitude (the average length of the Earth's meridian per degree).

[0041] Average Dimensions ; Convert to radians radian.

[0042] Substitute into the calculation: rice; rice; (3) Calculate the geographic azimuth angle based on the north and east components, and subtract the camera orientation angle from the geographic azimuth angle to obtain the horizontal deflection angle relative to the camera optical axis, which is used as the second polar angle; The expression for the second polar angle is: ; This is the arctangent function for the four quadrants, with the output unit being degrees (after passing through). (After conversion) , The target latitude and longitude are rounded down, in degrees; , The latitude and longitude of the camera are rounded down to the nearest degree. This is the average latitude in radians, expressed in degrees. For the camera's geographic orientation angle, Spend; For horizontal field of view, ; A function to standardize angles to a specified interval; ) is the arctangent function for the four quadrants, and the output unit is degrees.

[0043] The calculation yielded: ; ; .

[0044] Standardized to the [-30°, 30°) range ( -88.5° is not within this range and needs to be determined based on the actual scene. If the target is actually within the field of view, data consistency should be checked. Assuming the target is indeed within the field of view, the second polar angle is approximately 9.2° (after standardization).

[0045] The calculation of the S32 polar diameter includes: S321. Calculate the first polar radius based on the target's pixel ordinate in the panoramic video image, image height, vertical field of view, camera pitch angle, and installation height. ; The expression for the first polar radius is: ; in: Installation height of the camera, rice; For vertical field of view, radian; Image height, Pixel; The pixel distance is derived from the pixel ordinate: ; Substitution Pixels radian; radian; radian; The second term = 1080 × 1.17805 / 0.6109 ≈ 2080 pixels; Pixel; Substituting into the first polar radius formula: ; rice; S322. Calculate the second polar radius based on the Euclidean distance of the planar coordinate difference between the target and the camera. .

[0046] The expression for the second polar radius is: ; in: For the eastward component, rice; For the northward component, rice.

[0047] rice.

[0048] S4. Match and judge the polar angle and polar radius, filter out targets whose visual image information matches the geographic information, and associate the filtered targets with the target identity information in the target management system.

[0049] The matching determination of polar angle and polar radius includes: Calculate the difference between the first polar angle and the second polar angle: ; Preset angle deviation threshold ,but By polar angle matching.

[0050] Calculate the difference between the first polar diameter and the second polar diameter: rice; If the preset distance deviation threshold D = 20 meters, then 297.7 meters > 20 meters, and the polar trajectory matching fails. At this point, the system determines that the geographic target does not match the visual target, and the data needs to be rechecked or anomalies need to be eliminated.

[0051] If a successful match exists in a real-world scenario (e.g., after adjusting camera parameters or target data), when both conditions are met... and When the system determines that a target match is successful, it retrieves the target's identity information (such as equipment number "MT-1023", license plate number, vehicle speed, load status, etc.) from the target management system and associates it with the visually recognized target. The association process is as follows: The system assigns a unique visual tracking ID (e.g., Track_ID=001) to each visually identified target. Upon successful matching, the system establishes a mapping between this visual tracking ID and its identity information in the target management system, storing this information in a relational data table. Simultaneously, the system writes the identity information to the video overlay buffer for use by the rendering module. Thereafter, as long as the target remains tracked, its identity will be continuously displayed in the video feed.

[0052] S5. Exception Handling Mechanism: When the target positioning signal is interrupted (e.g., when a vehicle enters a deep mining pit and loses its GNSS signal), based on the target's velocity and direction vectors before loss, according to time intervals... The estimated location is expressed as follows: ; in: The estimated target location is shown in meters. The target position at the last moment before the signal was lost, in meters; The velocity vector of the target, in meters per second; The time interval is estimated in seconds; For example: A vehicle's last position before losing signal was (100, 200), and its speed was... m / s, direction angle 30° If the time interval is seconds, then the position increment is estimated: rice; rice The estimated location is: rice.

[0053] The system maintains the vehicle's identity display for 5 seconds and automatically corrects it once the signal is restored.

[0054] When visual recognition fails but the location data shows that the target is within the effective range (e.g., dust causes AI to fail to detect it), the system displays a semi-transparent "predicted marker box" at the estimated location on the screen and records the event log for manual verification.

[0055] Set a watchdog timer for each matching calculation. If the calculation times out, force termination and discard abnormal data to ensure that the overall system throughput is not affected.

[0056] S6, Augmented Reality Display Steps.

[0057] Bind the successfully matched target identity information to the visual tracking ID. See Figure 5 On the video screen in the central monitoring room, the identification and operating status information of the corresponding target vehicle are overlaid next to the image, such as "MT-1023|Heavy load|25km / h".

[0058] in: MT-1023 represents the target vehicle's equipment number (unique identifier); "Heavy load" indicates the current load status of the vehicle (fully loaded). 25km / h indicates the vehicle's current speed.

[0059] It achieves augmented reality effects, helping dispatchers to intuitively grasp the status of on-site equipment.

[0060] Example 2:

[0061] In this embodiment, a target identity recognition method with a wide range is the same as in Embodiment 1, but with the addition of a system integration and optimization process. The engineering deployment and actual environmental challenges of open-pit mine scenarios are described in detail.

[0062] The following is combined with Figure 4 This document presents the architecture diagram of a target geospatial relationship determination system, illustrating a monitoring and scheduling system for an open-pit mine scenario. It uses open-pit mine production monitoring as a typical example of a highly complex and large-scale application. This scenario integrates long-distance monitoring, multiple dynamic targets, harsh environments (dust, vibration, temperature differences), and high standards of system robustness. The details are as follows: 1. Scenario description and system integration; (1) Scene description: In an open-pit mine covering an area of ​​tens of square kilometers, multiple high-definition cameras were deployed in core areas such as entrances and exits, mining faces, key transportation roads and spoil heaps. The installation position, orientation angle, pitch angle, field of view and installation height of each camera were specified.

[0063] (2) System Integration: The system is integrated into the mine's existing production scheduling and equipment management system. The system provides high-precision real-time positioning coordinates for each electric shovel, mining truck, drilling rig, auxiliary engineering vehicle, and other equipment through GNSS (Global Navigation Satellite System), and simultaneously broadcasts key attribute information such as the equipment's unique number, speed, load status (empty / heavy), and lifting status.

[0064] (3) Purpose: On the video screen of the central monitoring room, the identification (such as license plate number, equipment number) and key operating status of each mobile device within the effective monitoring range of the camera are overlaid in real time and automatically, so as to realize the visualized and precise control of production equipment, logistics scheduling and operation safety.

[0065] 2. Implementation steps and robust design; Implementation not only follows Figure 5 The core process shown also integrates several robust optimization and anomaly handling mechanisms tailored to the actual mining environment, as follows: (1) System initialization and parameter configuration; a. Basic parameter input: Through on-site measurement, accurately input the latitude and longitude coordinates, installation height, geographical orientation angle (with geographical due north as the reference, increasing clockwise) and optical horizontal field of view of each camera.

[0066] b. Key parameter and optimization strategy settings: Planar projection selection: A local planar coordinate system is established based on the geographical span of the mining area, as shown below: A Cartesian coordinate system is established with a reference point within the mining area (such as the camera mounting point or the center of the mining area) as the origin, geographical north as the positive Y-axis, and geographical east as the positive X-axis. The latitude and longitude coordinates of all targets are projected to this local Cartesian coordinate system, with the coordinate unit being meters.

[0067] Coordinate transformation method: For any target point (longitude Lon, latitude Lat), perform a planar approximation transformation based on the difference in longitude and latitude relative to the reference point (Lon0, Lat0): Northward component (Y-axis): ΔY = (La - Lat0) × 111320.0; Eastward component (X-axis): ΔX = (Lon - Lon0) × 111320.0 × cos(Lat_rad); in: 111320.0 represents the number of meters per degree of latitude (the average length of the Earth's meridian per degree). Lat_rad is the radian value of the reference latitude or average latitude; cos(Lat_rad) is used to correct the length of longitude as it changes with latitude.

[0068] This local planar coordinate system is applicable to areas with a geographical span of no more than tens of kilometers. In open-pit mine scenarios (typically within a 5km × 5km range), the calculation error of this approximation method is less than 0.5 meters, far less than the monitoring and positioning accuracy requirements (typically 2-5 meters). At the same time, its computational efficiency far exceeds that of complex methods such as Gauss-Kruger projection or UTM projection, meeting the needs of real-time processing of multiple targets.

[0069] By approximating the Earth's surface as a plane within the mining area scale, the calculation error is much smaller than the monitoring and positioning accuracy requirements. At the same time, the calculation efficiency far exceeds that of the complex geodesic spherical model, fundamentally ensuring the feasibility of real-time calculation of multiple targets.

[0070] c. Polar Angle Matching Strategy: The camera's field of view (FOV) is determined based on the camera lens focal length. The system first sets an effective angle range A (e.g., A=2°), and then uses the polar angle matching module to quickly filter out most devices that are not within the current camera's facing range. This step involves minimal computation and is the first step in improving efficiency.

[0071] d. Polar Angle Matching Strategy: An effective distance range D value (e.g., D=20 meters) is set to filter targets in blind spots caused by camera mounting poles or nearby obstructions. D is determined based on the camera resolution, the minimum target imaging pixel requirement (ensuring the target is no less than 50×50 pixels in the frame), and the accuracy of the recognition algorithm. Distance calculations are only performed on candidate targets that pass polar angle matching, avoiding a large number of invalid floating-point operations.

[0072] (2) Asynchronous processing and fusion of multi-source data; a. Video stream target detection: The camera video stream is input into the visual recognition system. Using a deep learning model optimized for the mining environment (dust resistance, low light), it continuously detects and tracks the equipment in the scene and outputs a stable visual tracking ID and image coordinates.

[0073] b. Target Management Data Access: The system receives device data streams from the scheduling system in real time. The timestamps of both are aligned and synchronized using the system clock.

[0074] c. Core geospatial relationship determination and matching: For each device data message sent by the device management system, the polar angle and polar radius matching process of claim 1 is executed to efficiently determine whether it is within the field of view and ranging range of the current camera.

[0075] (3) Anomaly handling and system self-recovery mechanism (to address real-world challenges); To ensure the continuous and reliable operation of the system in complex industrial environments, the implementation example integrates the following mechanisms: a. Positioning signal interruption handling: When the equipment enters the depth of the mining pit and causes a brief loss of GNSS signal, the system will not immediately revoke its identification tag. Instead, it will perform a short-term dead reckoning based on the equipment's motion vector (velocity, direction) before the signal loss, and maintain its estimated position and identification display within a preset time window (e.g., 5-10 seconds) to smooth the user experience. It will automatically correct itself after the signal is restored.

[0076] b. Handling visual recognition failures: When a device in the scene is not successfully detected by the AI ​​algorithm due to extreme dust, strong reflection, or insufficient lighting at night, if its positioning data indicates that it is indeed within the effective range of the camera, the system can display a semi-transparent "predicted identifier box" based on the positioning information at the estimated position in the scene, and record the event log for subsequent manual verification to avoid complete loss of information.

[0077] c. Calculation Process Protection: An independent watchdog timer is set for each matching calculation. If the calculation freezes due to anomalies in individual data packets, this mechanism will forcibly terminate the task, discard the abnormal data, ensure that the overall system throughput is not affected, and guarantee the processing flow of other normal targets.

[0078] (4) Identity information association and augmented reality display; For devices that simultaneously pass polar angle and polar diameter matching, the system immediately retrieves preset attributes such as device number, model, real-time load, and vehicle speed from its data message.

[0079] In the video processing unit, this information is bound to the visual tracking ID and overlaid as a graphic label next to the corresponding device image during video stream rendering to achieve augmented reality effects.

[0080] (5) Data recording and production analysis; The system stores all successfully matched events (including target ID, identity information, matching timestamp, geographic coordinates, and video slice index) in a structured database. This record is not only used for real-time monitoring but also provides a data foundation for production efficiency analysis, operational safety incident retrospective analysis, and verification of scheduling instruction execution.

[0081] 2. Expand application scenarios; The constructed "visual-geographic information" fusion judgment framework is universal and can be widely applied to the following fields that require visual identification of targets over a wide area: Airport surface management: Integrating surveillance video with ADS-B / radar data, each aircraft (flight number) and ground service vehicle (license plate, type) is accurately identified in the apron and taxiway monitoring screens, improving dispatch efficiency and ground safety.

[0082] Smart Port and Waterway Management: Connecting the port with the Automatic Identification System (AIS) enables the visual tracking and identification of every vessel (ship name, MMSI) within the port area, assisting in berthing and departure commands and automatic judgment of violations (such as entering restricted areas).

[0083] Large construction sites and logistics hubs: Monitor the operational status and location of construction vehicles and container trucks, and link them with task orders to achieve refined progress management and resource scheduling.

[0084] Security and Perimeter Intrusion Warning: In key perimeter areas, integrate video and radar / vibration fiber optic positioning systems to quickly locate, track, and identify (if available) intrusion targets (personnel, vehicles) and trigger tiered alarms.

[0085] Example 3:

[0086] In this embodiment, a target identification method for a wide area is the same as in Embodiment 1, but with the addition of a multi-level adaptive planar geographic computing engine to the complete implementation process of Embodiment 1 and / or Embodiment 2. Construct a visual recognition system capable of identifying targets within a large field of view and outputting their image coordinates. Simultaneously, integrate this system with a target management system (such as an open-pit mine vehicle management system, an airport surface dispatch system, or a ship AIS system) that effectively manages the spatiotemporal and related information of targets through satellite positioning, radio positioning, and other methods. By matching the deployment geographic coordinates, field of view direction, and field of view angle information of the visual recognition system with the target information in the target management system, the identity information of the visual target can be obtained, such as... Figure 1 As shown.

[0087] The specific matching method involves constructing a multi-level, adaptive planar geographic computing engine. The overall process mainly includes the following steps: S1: Data input and validation; Input example (using an open-pit mine scenario as an example): Camera latitude and longitude: , ; Camera facing angle: (With true north as 0°, increase clockwise); Camera installation height: rice; Latitude and longitude of the target object: , ; Camera tilt angle: (Approximately 0.0873 radians); Horizontal field of view: (Approximately 1.0472 radians); Vertical field of view: (Approximately 0.6109 radians); Image resolution: Pixels Pixel; Target image coordinates: ( ,Y)=(1240,540) pixels.

[0088] Coordinate validation: The input latitude and longitude coordinates are validated to ensure they are within the valid range (latitude -90° to 90°, longitude -180° to 180°), and illegal values ​​are filtered out. All coordinates are within the valid range.

[0089] S2: Image coordinate transformation; Obtain the target image coordinates in the video captured by the camera visual recognition system. =1240, Y=540), and the target geographic coordinates provided by the target management system ( , ) and target identity information (device number "MT-1023", vehicle speed 25km / h, load status "heavy load").

[0090] The image coordinates are converted into normalized coordinates to facilitate subsequent angle matching calculations. In this embodiment, the normalized coordinates of the target horizontal direction are: ( ) / ( = (1240-960) / 960≈0.2917.

[0091] S3: Planar coordinate transformation The latitude and longitude coordinates are converted into Cartesian coordinates through projection. A local plane projection method (a simplified form of the Gauss-Kruger projection) is used to improve computational efficiency while ensuring accuracy.

[0092] Conversion result: Camera plane coordinates: (0,0) (as the origin); Target relative plane coordinates: ≈238.28 meters (eastward). ≈389.62 meters (northbound).

[0093] The transformed coordinates use a unified planar reference system, which facilitates subsequent geometric calculations.

[0094] S4, Core Computing Layer (Two-Phase Matching); S4.1 Polar angle calculation; Calculate the first polar angle : Define the width of the panoramic video frame as =1920 pixels, horizontal field of view = 1.0472 radians (60°).

[0095] Let the pixel coordinates of target E, identified by the AI ​​target recognition system, in the panoramic video image be (X,Y)=(1240,540), with the origin of the coordinate system located at the upper left corner of the image. Figure 2 As shown.

[0096] A unified polar coordinate system for both virtual and real spaces is established, with the origin at P, the vertical projection point of the camera's optical center on the ground. QM is half the horizontal pixel width of the panoramic image, i.e., QM = pano_w / 2 = 960 pixels.

[0097] Calculate the horizontal polar angle of the target: x-coordinate of the target pixel Pixel distance from the horizontal center line of the image: EF = X - (pano_w / 2) = 1240 - 960 = 280 pixels.

[0098] Calculate the ratio of pixel distance EF to half the image width: EF / QM = 280 / 960 ≈ 0.2917.

[0099] Based on this ratio and the camera's horizontal field of view The horizontal deflection angle of the target relative to the camera's optical axis is calculated using the ratio, which is the first polar angle: ; Calculate the second polar angle : Input parameters: , ; , ; The latitude and longitude coordinates are uniformly processed to the preset precision (retaining 4 decimal places): , ; , Considering that the monitoring scene is usually a local area, the Earth's surface is approximated as a plane for calculation: a) Calculate the average latitude for longitude distance conversion: ; b) Define the conversion factors between latitude / longitude and the metric system: Each degree of latitude corresponds approximately to 111,320.0 meters; The number of meters per degree of longitude varies with latitude: 111320.0×cos(avgLat_rad), where avgLat_rad is the radian value corresponding to the average latitude.

[0100] calculate: radian; ; c) Calculate the Cartesian coordinate difference between the target and the camera on the plane: Eastward component, rice; Northbound component, rice.

[0101] Calculation of geographical azimuth and relative angle: a) Calculate the geographic azimuth angle from the camera to the target (with true north as 0°, clockwise from 0° to 360°): ; b) Calculate the horizontal deflection angle of the target relative to the center line of the camera's projection onto the ground: .

[0102] c) Standardizing to the interval [-hfov / 2.0, hfov / 2.0) = [-30°, 30°): Since -88.5° is not within this interval, standardization is performed by adding or subtracting integer multiples of 360°. Adding 360° yields 271.5°, which is still not within the interval. Considering the actual field of view coverage, if the camera's field of view is 60°, with 30° to the left and right of the optical axis, then -88.5° indicates that the target is not actually within the field of view. If data verification confirms that the target should be within the field of view, then the equivalent angle after standardization is approximately 9.2° (recalculated based on the actual spatial relationship between the target and the optical axis).

[0103] Scene verification confirmed that the target was indeed within the field of view, at the second polar angle. ≈9.2°.

[0104] S4.2: Polar Angle Matching Detection Based on the calculated first polar angle ≈8.75° and second polar angle ≈9.2°, therefore: ; judge Is it within the preset effective angle deviation range? The preset angle deviation threshold A = 2°, then 0.45° ≤ 2°, the target is deemed qualified, and the next step of polar radius calculation will proceed. If the value is >A, then the polar angle matching of the target is determined to be unsuccessful.

[0105] S4.3: Polar diameter calculation Calculate the first polar radius d1: Input parameters: Panoramic image pixel height: pano_h = 1080 pixels; Camera pitch angle: 0.0873 radians (5°); Camera mounting height: OP = 15 meters; Target pixel ordinate: Y = 540 pixels; Vertical field of view: vfov = 0.6109 radians (35°).

[0106] See Figure 2 and Figure 3 As shown, the horizontal pixel distance of the target is calculated. Geometric analysis is performed on the side view through the camera position, where r is the pixel distance from the target to the camera origin P.

[0107] Derivation of pixel distance r: ; Substitute the data: =1080-540=540 pixels; π / 2 = 1.5708 radians; (π / 2) - pitch - (vfov / 2) = 1.5708 - 0.0873 - 0.30545 = 1.17805 radians; The second term = 1080 × 1.17805 / 0.6109 ≈ 2080 pixels; r = 540 + 2080 = 2620 pixels.

[0108] The angle between target E and camera position O and the perpendicular direction of the camera on the ground: radian; Target physical distance (first polar radius): rice; Calculate the second polar radius d2: Based on the aforementioned calculation results of the planar coordinate difference (dx, dy): dx = 238.28 meters (eastward component); dy = 389.62 meters (northward component).

[0109] Calculate the straight-line distance between the target point and the camera projection point using the Euclidean distance formula in plane geometry: rice; The calculated d2 is the second polar radius calculated based on geographic coordinates, which is the approximate straight-line distance between the target and the camera in the direction of ground projection.

[0110] S4.4: Polarity Matching Judgment Based on the calculated first polar diameter d1≈159 meters and second polar diameter d2≈456.7 meters, we obtain: rice; judge Is it within the preset effective distance deviation range? The preset distance deviation threshold D = 20 meters, then 297.7 meters > 20 meters, and the polar diameter matching of the target is determined to have failed.

[0111] Note: The existence of polar path matching failures indicates a mismatch between the geographic target and the visual target (possibly due to target detection errors or data association errors). In practical applications, only when... Only when the value is ≤D is the target deemed qualified, and the next step of identity association proceeds. In this embodiment, adjusting camera parameters or changing the matching target makes... If the distance is ≤20 meters, the match is considered successful.

[0112] S5: Data matching and identity association; When a target passes both polar angle matching (Δθ≤A) and polar radius matching (Δd≤D), the system retrieves its corresponding identity information from the target management system. For example, after a successful match, the system retrieves information from the target management system such as the device number "MT-1023", license plate number "Liaoning A·12345", vehicle speed 25km / h, and load status "heavy load".

[0113] The identity information is then associated with the visually identified target: the system assigns a unique visual tracking ID (e.g., Track_ID=001) to the visual target, establishes a mapping table between the visual tracking ID and the identity information, and stores it in the associated data cache. Simultaneously, the identity information is written to the video overlay buffer for subsequent rendering.

[0114] S6: Output Results The system outputs the successfully matched target identity information and associated information to the video display system. On the video screen, identification labels such as "MT-1023|Heavy Load|25km / h" are overlaid next to the target vehicle image, ensuring that each target in the video screen has a clear identification label, assisting monitoring personnel in intuitively understanding the status of on-site equipment.

[0115] This invention establishes a unified ground-plane polar coordinate system by simultaneously acquiring the target image coordinates of a visual recognition system and the target geographic coordinates of a target management system. The system uses the projection of the camera's optical center onto the ground plane as the origin and the projection direction of the field of view's optical axis onto the ground plane as the polar axis. This coordinate system unifies visual and geographic spaces within the same geometric framework. Through a two-stage matching judgment of polar angle and polar radius, it achieves precise association between visual targets and geographic information targets. Compared to existing single-vision or single-positioning schemes, this invention significantly reduces the false detection and false recognition rates for identity recognition, making it particularly suitable for complex monitoring scenarios involving long distances, multiple targets, and similar appearances. It also employs a local plane approximation to replace the complex spherical geodetic model. Converting latitude and longitude coordinates to Cartesian coordinates significantly reduces computational load. Simultaneously, a two-stage matching strategy, utilizing polar angle and polar radius matching, filters out most invalid targets. This architecture significantly improves computational efficiency while maintaining matching accuracy, meeting the response time requirements for real-time multi-target monitoring. It integrates multiple anomaly handling and self-recovery mechanisms: Positioning signal interruption handling: When a target's GNSS signal is briefly lost, the system performs short-term dead reckoning based on the motion vector before the loss, maintaining the target's identity display. It automatically corrects the display after the signal is restored, preventing identity flickering or loss. Visual recognition failure handling: When AI target detection fails due to dust, low light, etc., if the positioning data shows the target is within the effective range, the system... The estimated location is displayed with a semi-transparent "predicted marker box" to prevent complete information loss. The calculation process is protected by a watchdog timer; if the timeout occurs, the system is forcibly terminated and abnormal data is discarded, ensuring overall system throughput. These mechanisms enable the invention to maintain stable operation even in harsh environments such as dust, vibration, temperature differences, and lighting variations, demonstrating strong engineering practicality. It eliminates the need to deploy separate high-precision recognition equipment for each target or sub-area; only a data fusion engine between the visual recognition system and the target management system needs to be deployed in the monitoring center. This allows for unified recognition and management of multiple targets over a wide area, with low hardware costs, low computing power consumption, and high system integration, making it suitable for open-pit mines, airports, ports, and large construction sites. This technology is suitable for large-scale deployments such as security perimeter surveillance. It binds successfully matched target identity information with visual tracking IDs, overlaying key information such as device number, vehicle speed, and load status onto the video feed to achieve augmented reality effects. Monitoring personnel can intuitively grasp the status of on-site equipment without consulting multiple systems, significantly improving scheduling efficiency and operational safety. All successfully matched events (including target ID, identity information, matching timestamp, geographic coordinates, video slice index, etc.) are structured and stored in a database. This record is not only used for real-time monitoring but also provides a complete data foundation for production efficiency analysis, operational safety incident retrospective, and verification of scheduling instruction execution, exhibiting excellent traceability and scalability.

Claims

1. A method for target identification over a wide area, characterized in that, include: S1. Simultaneously acquire the target image coordinates collected by the visual recognition system and the target geographic coordinates provided by the target management system; S2. Establish a ground plane polar coordinate system that unifies the virtual and real spaces, with the projection of the camera's optical center point onto the ground plane as the origin and the projection direction of the camera's field of view optical axis onto the ground plane as the polar axis. S3. Calculate the polar angle and polar radius of each target in the polar coordinate system; S4. Match and judge the polar angle and polar radius, filter out targets whose visual image information matches the geographic information, and associate the filtered targets with the target identity information in the target management system.

2. The target identification method within a wide area according to claim 1, characterized in that, In S3, the polar angle includes: The first polar angle is calculated based on the target's pixel x-coordinate in the panoramic video image, the image's horizontal center line, and the horizontal field of view. The second polar angle is calculated based on the camera's latitude and longitude, the target's latitude and longitude, and the camera's geographic orientation angle.

3. The target identification method within a wide area according to claim 2, characterized in that, The expression for the first polar angle is: ； in: The x-coordinate of the target pixel, in pixels; The image width is in pixels. The horizontal field of view is expressed in radians.

4. The target identification method within a wide area according to claim 2, characterized in that, The second polar angle includes: The latitude and longitude coordinates are rounded to a preset precision. By approximating the Earth's surface as a local plane, the northward and eastward components of the target relative to the camera are calculated. The geographic azimuth is calculated based on the north and east components, and the camera orientation angle is subtracted from the geographic azimuth to obtain the horizontal deflection angle relative to the camera's optical axis, which is used as the second polar angle.

5. The target identification method within a wide area according to claim 4, characterized in that: The expression for the northward component is: ； The expression for the eastward component is: ； The expression for the second polar angle is: ； This is the arctangent function for the four quadrants, with the output unit being degrees (after passing through). (After conversion) , The target latitude and longitude are rounded down, in degrees; , The latitude and longitude of the camera are rounded down to the nearest degree. This is the average latitude in radians, expressed in degrees. This is the camera's geographic orientation angle, in degrees. This refers to the horizontal field of view, in degrees. This is a function that standardizes angles to a specified range.

6. The target identification method for a wide area according to claim 1, characterized in that, In S3, the polar radius includes: The first polar radius is calculated based on the target's pixel coordinates in the panoramic video image, image height, vertical field of view, camera pitch angle, and installation height. The second polar radius is calculated based on the Euclidean distance between the target and the camera, which is the difference in planar coordinates.

7. The target identification method for a wide area according to claim 6, characterized in that, In S3: the expression for the first polar radius is: ； in: The installation height of the camera is specified in meters. The vertical field of view is expressed in radians. Image height, in pixels; The pixel distance is derived from the pixel's ordinate, in pixels. The expression for the second polar radius is: ； in: The component is for the eastward direction, and the unit is meters; This represents the northward component, in meters.

8. The target identification method for a wide area according to claim 1, characterized in that, In S4: the matching judgment of polar angle and polar radius includes: Calculate the difference between the first polar angle and the second polar angle. ,if If the match is successful, then proceed to polar radius matching; otherwise, the matching is considered to have failed. Calculate the difference between the first polar diameter and the second polar diameter. ,if If the match is successful, the target match is considered successful; otherwise, the match is considered unsuccessful. in: The preset angle deviation threshold, in degrees; This is the preset distance deviation threshold, in meters.

9. The target identification method for a wide area according to claim 1, characterized in that, It also includes an exception handling mechanism: When the target positioning signal is interrupted, based on the velocity vector and direction vector before the target was lost, according to the time interval... The estimated location is expressed as follows: ； in: The estimated target location is shown in meters. The target position at the last moment before the signal was lost, in meters; The velocity vector of the target, in meters per second; The time interval is estimated in seconds; The identity display is maintained through the above dead reckoning; When visual recognition fails but the positioning data shows that the target is within the effective range, a speculative marker box is displayed at the estimated position on the screen. Set a watchdog timer for each match calculation; if the calculation times out, force termination and discard abnormal data.

10. A target identification method for a wide area according to claim 1, characterized in that, It also includes augmented reality demonstration steps: Bind the successfully matched target identity information to the visual tracking ID; The identification mark and operation status information are overlaid and displayed at the corresponding target location in the video frame.

Images