AI-based unmanned aerial vehicle dynamic target tracking system and method

Abstract

The application discloses an unmanned aerial vehicle dynamic target tracking system and method based on AI, and particularly relates to the technical field of intelligent unmanned aerial vehicles, a system startup initialization module, a system main control module, an AI visual positioning acquisition module, a tracking range monitoring module, a hierarchical early warning prompt module, a position recognition and sound-light warning module, a global obstacle perception module, an AI multi-path intelligent planning module, a flight dynamic regulation module, and a parking pretreatment module; aiming at the short board of complex airspace operation of intelligent unmanned aerial vehicles, a full-process safety protection system of grid obstacle density weighted detection, multi-dimensional intelligent path optimization, and dynamic flight attitude regulation is constructed; through laser radar and visual dual-sensor fusion detection, combined with obstacle hardness and motion state, the obstacle comprehensive density is weighted calculated, and the airspace danger level is accurately quantified; meanwhile, the obstacle density, airspace wind resistance, and flight time are taken as core dimensions to complete multi-path intelligent screening, and multiple optimal alternative routes are reserved and artificial optimization is supported.

Description

Technical Field

[0001] This invention relates to the field of intelligent unmanned aerial vehicle (UAV) technology, and more specifically, to an AI-based UAV dynamic target tracking system and method. Background Technology

[0002] In the field of intelligent unmanned aerial vehicle manufacturing technology, UAV dynamic target tracking technology is widely used in various low-altitude operation scenarios such as security inspection, dynamic monitoring, and on-site duty.

[0003] Currently, most conventional intelligent UAV tracking systems on the market adopt fixed threshold judgment logic and single visual recognition schemes, resulting in a low overall level of intelligence and numerous technical shortcomings in actual operation. Traditional tracking methods rely solely on single-frame visual data to complete target localization and positioning determination, leading to poor recognition stability, weak anti-interference capabilities, and susceptibility to tracking errors and target loss caused by ambient light and airspace clutter. Traditional UAV obstacle avoidance route planning is limited to a single path, lacking multi-path optimization and manual alternative mechanisms, making it unable to adapt to dynamic conditions such as airspace wind resistance and terrain changes. Furthermore, flight attitude control is lagging, making it difficult to cope with sudden situations such as airflow disturbances and temporary obstacle changes. The overall stability, safety, and intelligence level of tracking operations are insufficient to meet the current requirements for the mass production and practical application of high-precision and high-reliability intelligent UAVs. Therefore, we propose an AI-based UAV dynamic target tracking system and method. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, the present invention provides an AI-based dynamic target tracking system and method for unmanned aerial vehicles (UAVs) to address the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an AI-based drone dynamic target tracking system, comprising: The system startup initialization module, as the core of the system startup, is used to receive external power-on commands, complete the system-wide hardware self-test, parameter initialization and communication pre-configuration of each sub-module, and output the system startup ready signal; The system main control module receives the ready signal from the system startup initialization module and establishes a full-domain control link. As the core control hub of the system, it is responsible for coordinating the receipt of various sensor data, running AI algorithm logic, issuing control and early warning commands in a hierarchical manner, and coordinating the whole machine tracking operation process. The AI ​​visual positioning and acquisition module establishes a real-time communication link with the system's main control module. It is used to lock onto dynamically tracked targets by relying on the airborne visual recognition unit, collect the target's three-dimensional spatial coordinates and displacement trajectory information in real time, and simultaneously acquire the UAV's real-time flight coordinates, flight attitude, and basic airspace environment data. The tracking range monitoring module has two levels of height thresholds and two levels of distance thresholds. The two levels of height thresholds include a first height threshold and a second height threshold. The two levels of distance thresholds include a first distance threshold and a second distance threshold. It is used to calculate the relative height difference and straight-line tracking distance between the UAV and the dynamic target in real time, and continuously compare the measured data with the preset thresholds to determine whether the UAV tracking operation has deviated from the safe and controllable range. The graded early warning module is electrically connected to the tracking range monitoring module and the system main control module. It is used to execute graded early warning logic according to the degree of tracking range exceeding the limit. When the first-level threshold is exceeded, a background pop-up warning is output and the flight attitude is automatically fine-tuned. When the second-level threshold is exceeded, an airborne buzzer warning is activated at the same time, and the over-limit flight displacement of the UAV is restricted. The positioning recognition and audio-visual warning module is connected to the system's main control module. It is used to verify the drone's tracking and positioning status through multiple frames of continuous position data. When it is determined that the drone has reached the preset target tracking point and the tracking task has been completed, it adaptively adjusts the audio-visual intensity to activate the onboard alarm and attracts the attention of people in the surrounding area through audio-visual signals. The all-domain obstacle perception module, with built-in lidar and high-definition camera sensor unit, is used to scan the drone's flight airspace in a grid pattern, collect data on the distribution location, size, and motion status of surrounding obstacles, and calculate the comprehensive distribution density of obstacles by combining obstacle hardness parameters. The AI ​​multi-path intelligent planning module communicates bidirectionally with the all-domain obstacle perception module and the system main control module. When the obstacle density exceeds the preset critical value, it combines multiple parameters such as airspace wind force, geographical topography, obstacle distribution, and flight time to generate multiple feasible navigation paths. It also intelligently filters and scores the multiple paths and pushes the optimal alternative path set to the control terminal for the user to choose from. The flight dynamic control module receives output commands from the system's main control module and the AI ​​multi-path intelligent planning module. It is used to monitor changes in the airspace environment in real time according to the selected navigation path, dynamically fine-tune the UAV's flight attitude, cruise speed and flight path, and avoid environmental damage such as obstacle collisions and airflow impacts. The shutdown preprocessing module is used to receive shutdown termination instructions from the system's main control module and complete pre-shutdown preparations such as backing up tracking operation data, archiving flight parameters, and resetting airborne actuators.

[0006] Preferably, the two-level thresholds of the tracking range monitoring module are calibrated through AI big data fitting training. By collecting historical data on the long-term activity radius of the target and the terrain undulations of the tracking area, threshold parameters adapted to the dynamic target movement characteristics are generated. The first height threshold and the first distance threshold are small fluctuation warning thresholds, and the second height threshold and the second distance threshold are safety boundary mandatory alarm thresholds.

[0007] Preferably, the tracking range monitoring module calculates the relative height difference and straight-line tracking distance between the UAV and the dynamic target in real time, and combines this with preset two-level height thresholds and two-level distance thresholds to determine whether the tracking operation exceeds the safe and controllable range. The tracking range monitoring module has a built-in precise distance calculation formula, which is used to quantify the straight-line tracking distance between the UAV and the dynamic target in real time, providing data basis for threshold comparison and safe range determination. The formula is as follows:

[0008] In the formula, For real-time linear tracking of the distance between the drone and the dynamic target; , , () represents the real-time three-dimensional coordinates of the UAV; , , () represents the real-time three-dimensional coordinates of a dynamic target; calculated in real time. The system matches the preset first distance threshold and second distance threshold, and uses the real-time height difference to determine the safe tracking range.

[0009] Preferably, the positioning recognition and audio-visual warning module's positioning determination logic is to continuously collect no less than 5 frames of target position data, and only when the UAV's three-dimensional coordinates continuously fall within the preset point error range and the AI ​​predicts that the target has no large escape displacement, is the tracking and positioning determined; the audio-visual warning module adaptively adjusts the audio-visual output power according to the ambient light intensity and noise decibels.

[0010] Preferably, the all-domain obstacle perception module adopts a cubic grid detection mechanism, dividing the airspace surrounding the UAV into several monitoring grids. By fusing LiDAR point cloud data with image semantic recognition data, the module calculates the spatial proportion of obstacles in each grid, and combines this with a weighted density calculation based on obstacle attributes to accurately determine the airspace congestion level. The comprehensive obstacle density is calculated using a simplified weighted formula, as follows:

[0011] In the formula, This represents the overall distribution density of obstacles within the grid. This represents the percentage of the space occupied by the obstacle within the detection grid. This is a weighted coefficient for obstacle hardness. The coefficient for hard obstacles is greater than that for soft obstacles, and the coefficient for stationary obstacles is less than that for moving obstacles. This weighted correction improves the accuracy of density detection.

[0012] Preferably, the path selection dimensions of the AI ​​multi-path intelligent planning module include three core indicators: obstacle density, airspace wind resistance, and flight time. It eliminates inferior routes through a multi-dimensional scoring mechanism, retains 3-5 comprehensive optimal routes, and simultaneously labels the estimated flight time, risk type, and recommended cruise speed parameters for each route.

[0013] Preferably, the flight dynamic control module monitors changes in the airspace environment in real time along the selected navigation path, and makes small adjustments to the UAV's pitch angle, roll angle, and flight speed in real time in response to sudden airflow disturbances or temporary obstacle changes.

[0014] The AI-based method for dynamic target tracking of drones includes the following steps: Step 1: The system startup initialization module receives the external startup command, completes the whole machine hardware self-test, algorithm parameter initialization and communication configuration of each sub-module, and sends a startup ready signal to the system main control module; Step 2: The system main control module activates the full-domain communication link and sequentially completes the communication connection of the AI ​​visual positioning and acquisition module, the tracking range monitoring module, the full-domain obstacle perception module, and the AI ​​multi-path intelligent planning module. The system then enters standby tracking state. Step 3: The AI ​​visual positioning and acquisition module starts dynamic target recognition, collects target spatial coordinates, displacement trajectory and UAV flight status data in real time, and transmits them synchronously to the system main control module and tracking range monitoring module; Step 4: The tracking range monitoring module calculates the relative height difference and straight-line tracking distance between the UAV and the target in real time, and compares them with the preset two-level height threshold and distance threshold. If the measured data exceeds any level threshold, the graded early warning module is triggered to execute the corresponding early warning action and fine-tune the UAV's flight attitude to return to the safe range. Step 5: The system continuously verifies the tracking point through the positioning recognition and sound and light warning modules. After verifying the tracking position and the target status is stable through multiple frames of data, the system adaptively activates the airborne alarm to provide on-site warning and attract the attention of the surrounding area through sound and light signals. Step Six: During flight, the all-domain obstacle perception module continuously scans the surrounding airspace in a grid pattern, calculates the overall obstacle density with weights, and activates the AI ​​multi-path intelligent planning module when the density exceeds the preset threshold. Step 7: The AI ​​multi-path intelligent planning module combines airspace environment, obstacle distribution, and wind resistance parameters to deduce multiple alternative routes. After multi-dimensional scoring and screening, the routes are pushed to the control terminal, where the user selects the optimal flight path. Step 8: The flight dynamic control module monitors the dynamic changes in the airspace environment in real time according to the path selected by the user, and continuously fine-tunes the flight attitude and route of the UAV to avoid environmental interference such as obstacles and airflow, and protect the safety of the UAV as a whole. Step Nine: After the tracking operation is completed or a shutdown command is received, the shutdown preprocessing module completes preliminary work such as data backup and mechanism reset. After the shutdown status verification module verifies that the shutdown conditions meet the standards, the system closing module completes log archiving, data verification and resource release, and the system stops running.

[0015] The technical effects and advantages of this invention are as follows: This invention addresses the shortcomings of intelligent unmanned aerial vehicles (UAVs) operating in complex airspaces by constructing a comprehensive safety protection system encompassing grid-based obstacle density weighted detection, multi-dimensional intelligent path optimization, and dynamic flight attitude control. Through the fusion of LiDAR and visual sensors, and combining obstacle hardness and motion state to calculate the comprehensive obstacle density, it accurately quantifies the airspace hazard level, providing precise data support for path planning. Simultaneously, it performs multi-path intelligent screening based on obstacle density, airspace wind resistance, and flight time as core dimensions, retaining multiple optimal alternative routes and supporting manual selection, overcoming the shortcomings of traditional aircraft's single obstacle avoidance path and incomplete risk avoidance. Coupled with a real-time environmental perception-based dynamic attitude and speed fine-tuning mechanism, it can quickly adapt to sudden conditions such as airflow disturbances and temporary obstacle changes, greatly improving the stability and safety of UAVs flying in complex airspaces, reducing the probability of collisions and airflow impact damage, effectively extending the overall service life of intelligent UAVs, and enhancing the intelligent operation level and scenario adaptability of aircraft, thus meeting the upgrading and development needs of the high-end intelligent UAV manufacturing industry.

[0016] When in use, this invention solves the industry pain points of poor threshold adaptability, large ranging error, and easy false triggering of positioning warnings in traditional aircraft tracking. It significantly improves the positioning accuracy, safety judgment reliability, and environmental anti-interference capability of UAV dynamic target tracking. Through a graded early warning linkage attitude fine-tuning and displacement limit mechanism, it realizes graded risk management and avoids safety hazards such as tracking overstepping boundaries and flight loss of control from the source. It significantly improves the standardization and accuracy of intelligent UAV dynamic tracking operations and adapts to the operational needs of various complex dynamic target tracking scenarios. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall structure of the present invention.

[0018] Figure 2 This is a flowchart of the structural method of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] As attached Figure 1 The AI-based drone dynamic target tracking system shown includes: The system startup initialization module, as the core of the system startup, is used to receive external power-on commands, complete the system-wide hardware self-test, parameter initialization and communication pre-configuration of each sub-module, and output the system startup ready signal; The system main control module receives the ready signal from the system startup initialization module and establishes a full-domain control link. As the core control hub of the system, it is responsible for coordinating the receipt of various sensor data, running AI algorithm logic, issuing control and early warning commands in a hierarchical manner, and coordinating the whole machine tracking operation process. The AI ​​visual positioning and acquisition module establishes a real-time communication link with the system's main control module. It is used to lock onto dynamically tracked targets by relying on the airborne visual recognition unit, collect the target's three-dimensional spatial coordinates and displacement trajectory information in real time, and simultaneously acquire the UAV's real-time flight coordinates, flight attitude, and basic airspace environment data. The tracking range monitoring module has two levels of height thresholds and two levels of distance thresholds. The two levels of height thresholds include a first height threshold and a second height threshold. The two levels of distance thresholds include a first distance threshold and a second distance threshold. It is used to calculate the relative height difference and straight-line tracking distance between the UAV and the dynamic target in real time, and continuously compare the measured data with the preset thresholds to determine whether the UAV tracking operation has deviated from the safe and controllable range. The graded early warning module is electrically connected to the tracking range monitoring module and the system main control module. It is used to execute graded early warning logic according to the degree of tracking range exceeding the limit. When the first-level threshold is exceeded, a background pop-up warning is output and the flight attitude is automatically fine-tuned. When the second-level threshold is exceeded, an airborne buzzer warning is activated at the same time, and the over-limit flight displacement of the UAV is restricted. The positioning recognition and audio-visual warning module is connected to the system's main control module. It is used to verify the drone's tracking and positioning status through multiple frames of continuous position data. When it is determined that the drone has reached the preset target tracking point and the tracking task has been completed, it adaptively adjusts the audio-visual intensity to activate the onboard alarm and attracts the attention of people in the surrounding area through audio-visual signals. The all-domain obstacle perception module, with built-in lidar and high-definition camera sensor unit, is used to scan the drone's flight airspace in a grid pattern, collect data on the distribution location, size, and motion status of surrounding obstacles, and calculate the comprehensive distribution density of obstacles by combining obstacle hardness parameters. The AI ​​multi-path intelligent planning module communicates bidirectionally with the all-domain obstacle perception module and the system main control module. When the obstacle density exceeds the preset critical value, it combines multiple parameters such as airspace wind force, geographical topography, obstacle distribution, and flight time to generate multiple feasible navigation paths. It also intelligently filters and scores the multiple paths and pushes the optimal alternative path set to the control terminal for the user to choose from. The flight dynamic control module receives output commands from the system's main control module and the AI ​​multi-path intelligent planning module. It is used to monitor changes in the airspace environment in real time according to the selected navigation path, dynamically fine-tune the UAV's flight attitude, cruise speed and flight path, and avoid environmental damage such as obstacle collisions and airflow impacts. The shutdown preprocessing module is used to receive the shutdown termination command from the system's main control module and complete pre-shutdown preparations such as backing up tracking operation data, archiving flight parameters, and resetting airborne actuators. The two-level thresholds of the tracking range monitoring module are calibrated through AI big data fitting training. By collecting historical data on the long-term activity radius of the target and the terrain undulations of the tracking area, threshold parameters adapted to the dynamic target's motion characteristics are generated. The first height threshold and the first distance threshold are warning thresholds for small fluctuations, while the second height threshold and the second distance threshold are mandatory alarm thresholds for safety boundaries. This achieves adaptive calibration of tracking safety thresholds, overcoming the limitations of traditional fixed thresholds. It significantly reduces the probability of false alarms and missed alarms in tracking warnings, making the safety judgment rules more consistent with the real motion patterns of dynamic targets and more adaptable. The tracking range monitoring module calculates the relative height difference and straight-line tracking distance between the UAV and the dynamic target in real time, and combines this with preset two-level height thresholds and two-level distance thresholds to determine whether the tracking operation exceeds the safe and controllable range. The tracking range monitoring module has a built-in formula for accurate distance calculation, which is used to quantify the straight-line tracking distance between the UAV and the dynamic target in real time, providing data for threshold comparison and safe range determination. The formula is as follows:

[0021] In the formula, For real-time linear tracking of the distance between the drone and the dynamic target; , , () represents the real-time three-dimensional coordinates of the UAV; , , () represents the real-time three-dimensional coordinates of a dynamic target; calculated in real time. It matches preset first and second distance thresholds and uses real-time height difference to determine the tracking safety range; it quantifies tracking distance parameters to provide accurate data support for safety range determination; it solves the problem of large errors in traditional visual estimation and distance measurement, improves the real-time performance and accuracy of tracking boundary violation determination, and ensures reliable triggering of graded warnings; The positioning recognition and audible-visual warning module's positioning determination logic involves continuously collecting no less than 5 frames of target position data. Positioning is determined only when the UAV's 3D coordinates continuously fall within the preset position error range and the AI ​​predicts no significant escape displacement. The audible-visual warning module adaptively adjusts its output power based on ambient light intensity and noise levels. A multi-frame continuous position data cross-verification mechanism is employed, combined with target displacement prediction, to complete positioning determination. The audible-visual output intensity is adaptively adjusted based on ambient light intensity and noise levels. This achieves high-precision, anti-false-judgment positioning recognition and adaptive audible-visual warnings. It avoids false alarms caused by single-frame recognition jitter, adapts to various complex environmental scenarios, and significantly improves on-site warning recognition and operational safety. The all-domain obstacle perception module employs a cubic grid detection mechanism, dividing the airspace surrounding the drone into several monitoring grids. By fusing LiDAR point cloud data with image semantic recognition data, it calculates the spatial proportion of obstacles within each grid and combines this with a weighted density calculation based on obstacle attributes to accurately determine the airspace congestion level. The overall obstacle density is calculated using a simplified weighted formula, as follows:

[0022] In the formula, This represents the overall distribution density of obstacles within the grid. This represents the percentage of the space occupied by the obstacle within the detection grid. It is a weighted coefficient for obstacle hardness, with the coefficient for hard obstacles being greater than that for soft obstacles and the coefficient for stationary obstacles being less than that for moving obstacles. The accuracy of density detection is improved through weighted correction; the level of airspace obstacle congestion is quantified to accurately identify complex and dangerous flight areas; it can distinguish the danger weight of different obstacles, solve the problem of low accuracy of traditional single obstacle identification, and provide a reliable risk basis for intelligent path planning. The AI ​​multi-path intelligent planning module's path selection dimensions include three core indicators: obstacle density, airspace wind resistance, and flight time. It uses a multi-dimensional scoring mechanism to eliminate substandard routes, retaining 3-5 optimal routes. Each route is simultaneously labeled with estimated flight time, risk type, and recommended cruise speed parameters. Using obstacle density, airspace wind resistance, and flight time as core evaluation indicators, it quantitatively scores and selects the best from multiple candidate routes. It automatically selects multiple optimal safe routes and labels their operating parameters, supporting manual selection. It overcomes the limitations of traditional single-path obstacle avoidance, balancing flight safety and tracking efficiency, enhancing human-machine collaboration, and effectively avoiding flight risks in complex airspaces. The flight dynamic control module monitors changes in the airspace environment in real time along the selected navigation path. In response to sudden airflow disturbances or temporary obstacle movements, it makes small adjustments to the UAV's pitch angle, roll angle, and flight speed in real time. It also monitors the dynamic changes in the airspace environment in real time and dynamically adjusts the UAV's pitch angle, roll angle, and flight speed in response to airflow disturbances or temporary obstacle movements. This effectively suppresses flight deviation caused by environmental interference, improves the stability of tracking flight, and reduces the probability of UAV collisions, instability, and damage. like Figure 2 As shown, the AI-based drone dynamic target tracking method provided in this embodiment includes the following steps: Step 1: The system startup initialization module receives the external startup command, completes the whole machine hardware self-test, algorithm parameter initialization and communication configuration of each sub-module, and sends a startup ready signal to the system main control module; Step 2: The system main control module activates the full-domain communication link and sequentially completes the communication connection of the AI ​​visual positioning and acquisition module, the tracking range monitoring module, the full-domain obstacle perception module, and the AI ​​multi-path intelligent planning module. The system then enters standby tracking state. Step 3: The AI ​​visual positioning and acquisition module starts dynamic target recognition, collects target spatial coordinates, displacement trajectory and UAV flight status data in real time, and transmits them synchronously to the system main control module and tracking range monitoring module; Step 4: The tracking range monitoring module calculates the relative height difference and straight-line tracking distance between the UAV and the target in real time, and compares them with the preset two-level height threshold and distance threshold. If the measured data exceeds any level threshold, the graded early warning module is triggered to execute the corresponding early warning action and fine-tune the UAV's flight attitude to return to the safe range. Step 5: The system continuously verifies the tracking point through the positioning recognition and sound and light warning modules. After verifying the tracking position and the target status is stable through multiple frames of data, the system adaptively activates the airborne alarm to provide on-site warning and attract the attention of the surrounding area through sound and light signals. Step Six: During flight, the all-domain obstacle perception module continuously scans the surrounding airspace in a grid pattern, calculates the overall obstacle density with weights, and activates the AI ​​multi-path intelligent planning module when the density exceeds the preset threshold. Step 7: The AI ​​multi-path intelligent planning module combines airspace environment, obstacle distribution, and wind resistance parameters to deduce multiple alternative routes. After multi-dimensional scoring and screening, the routes are pushed to the control terminal, where the user selects the optimal flight path. Step 8: The flight dynamic control module monitors the dynamic changes in the airspace environment in real time according to the path selected by the user, and continuously fine-tunes the flight attitude and route of the UAV to avoid environmental interference such as obstacles and airflow, and protect the safety of the UAV as a whole. Step 9: After the tracking operation is completed or a shutdown command is received, the shutdown preprocessing module completes preliminary work such as data backup and mechanism reset. After the shutdown status verification module verifies that the shutdown conditions meet the standards, the system closing module completes log archiving, data verification and resource release, and the system stops running. The system completes system initialization, data acquisition, safety assessment, early warning and control, obstacle avoidance planning, dynamic flight control, and shutdown and cleanup through a step-by-step closed-loop process; it establishes a standardized and automated intelligent tracking operation process; and achieves intelligent and unmanned tracking operation throughout the entire process. The system's operating logic is stable and closed-loop, with high environmental adaptability, which greatly improves the overall reliability and safety of UAV dynamic tracking.

[0023] By combining a modular hardware collaborative architecture with AI intelligent algorithm logic, this system systematically solves many industry shortcomings in traditional UAV dynamic target tracking, such as poor adaptability of fixed safety thresholds, lack of graded early warning mechanisms for tracking out-of-bounds violations, susceptibility to collision damage in complex obstacle airspace, limited obstacle avoidance routes that cannot be manually optimized, flight attitude that cannot adapt to dynamic environmental changes, easy false triggering of positioning warnings, and low accuracy of environmental risk identification. The overall technical solution achieves fully automated target identification, intelligent safety assessment, graded risk early warning, precise obstacle detection, diversified path planning, dynamic flight control, and standardized shutdown processing, greatly improving the intelligence level, environmental adaptability, and flight safety of UAV dynamic tracking operations. When the system is running, the system startup initialization module first receives the external power-on command, completes the whole-domain hardware self-check, initialization of operating parameters, and pre-configuration of communication for each sub-module, and outputs a system startup ready signal after confirming that the whole machine is fault-free, providing a stable hardware foundation and communication guarantee for subsequent full-function operation. Then, the system main control module receives the startup ready signal and establishes a whole-domain control and data communication link. As the core control center of the whole machine, it uniformly coordinates the reception of various sensor data, the operation of AI algorithm logic, and the issuance of hierarchical control commands and early warning commands, and manages the entire process of UAV tracking operation. In the normal standby operation state of the system, the AI ​​vision positioning and acquisition module maintains real-time communication with the system main control module. Relying on the airborne vision recognition unit, it accurately locks the dynamic tracking target in the airspace, continuously collects the target's real-time three-dimensional spatial coordinates and displacement trajectory change information, and simultaneously acquires the UAV's own real-time flight coordinates, flight attitude parameters, and basic airspace environmental data, providing high-precision raw data support for subsequent tracking range calculation, obstacle identification, and route planning. To ensure that the tracking operation remains within a safe and controllable range, the tracking range monitoring module pre-stores two levels of height thresholds and two levels of distance thresholds. The two levels of height thresholds include a first height threshold and a second height threshold, and the two levels of distance thresholds include a first distance threshold and a second distance threshold. These thresholds are not traditional fixed parameters, but are calibrated and generated through AI big data fitting training. By collecting historical operational data such as the dynamic target's activity radius and the terrain undulations of the tracking area over a long period of time, and combining it with AI fitting algorithms, exclusive parameters adapted to the target's movement characteristics are obtained. The first height threshold and the first distance threshold are set as warning thresholds adapted to normal small position fluctuations, and the second height threshold and the second distance threshold are set as mandatory alarm thresholds for tracking out of the controllable range, effectively reducing the false alarms and missed alarms caused by traditional fixed thresholds. Meanwhile, the tracking range monitoring module relies on the built-in three-dimensional spatial ranging calculation formula to accurately calculate the straight-line tracking distance by inputting the three-dimensional coordinate data of the UAV and the dynamic target in real time. Combined with the real-time calculated relative height difference between the UAV and the target, it compares the data with two preset thresholds to accurately determine whether the current tracking operation has deviated from the safe and controllable range. When the tracking range exceeds the limit, the graded early warning module executes differentiated control logic according to the level of the violation. If only the first-level threshold is exceeded, the system background will automatically pop up an early warning and make slight adjustments to the UAV's flight attitude to try to bring the UAV back to the safe tracking range. When the second-level safety boundary threshold is exceeded, the system will continuously alarm in the background and activate the onboard buzzer warning, and restrict the UAV from continuing to move into the dangerous area, thus realizing graded risk control. During the continuous tracking of dynamic targets by the drone, the positioning recognition and sound and light warning modules continuously verify the tracking positioning status. The judgment logic adopts cross-verification of no less than five frames of continuous position data. Only when the three-dimensional coordinates of the drone continuously and stably fall within the preset position error range, and the AI ​​predicts that the target will not make a large escape displacement in a short period of time, can the tracking task be judged to be completed. Subsequently, the system adaptively adjusts the sound and light output power of the airborne alarm according to the ambient light intensity and ambient noise decibels. Through adaptive sound and light signals, it effectively attracts the attention of surrounding personnel, greatly improving the on-site warning effect and operational safety. For complex obstacle-filled airspace scenarios in the wild and densely populated urban areas, this invention achieves high-precision environmental risk identification through a global obstacle perception module. This module is equipped with a LiDAR and high-definition camera sensing unit, employing a cubic grid detection mechanism to divide the airspace surrounding the UAV into regions. It integrates LiDAR point cloud data and image semantic recognition results to calculate the proportion of obstacle space occupied within each grid. Simultaneously, it sets differentiated weighting coefficients based on obstacle hardness attributes and dynamic / static motion states, calculating the comprehensive obstacle distribution density using a simplified weighting formula. This accurately quantifies the degree of airspace congestion and hazard level, effectively distinguishing the damage risk to the UAV from different types of obstacles, and providing precise judgment for subsequent path planning. The AI ​​multi-path intelligent planning module is activated immediately when the overall density of detected obstacles exceeds the system's preset threshold. Combining multi-dimensional environmental parameters such as airspace wind conditions, geographical features, obstacle distribution, and flight time, it intelligently generates multiple safe navigation routes. It then uses obstacle density, airspace wind resistance, and flight time as core evaluation dimensions to quantitatively score all candidate routes, automatically eliminating high-risk and inefficient routes and selecting three to five optimal alternative routes. At the same time, it labels each route with key parameters such as estimated flight time, road segment risk type, and recommended cruising speed, and pushes these parameters to the control terminal for users to choose from. After the user selects the optimal navigation path, the flight dynamic control module tracks the flight path status throughout the entire process, monitors the dynamic changes in the airspace environment in real time, and fine-tunes the UAV's pitch angle, roll angle, and flight cruise speed in real time for sudden airflow disturbances, temporary obstacle movements, and other sudden and complex situations. It dynamically adapts to the real-time flight environment to minimize the risk of damage to the aircraft caused by obstacle collisions and airflow impacts. When the tracking operation is completed or an external shutdown termination command is received, the shutdown preprocessing module responds promptly, completing preparatory work such as backing up historical tracking data, archiving flight operation parameters, and resetting various onboard actuators. It works in conjunction with the closed-loop logic of the entire aircraft to complete the operation and ensure the stability and standardization of the system's start-up and shutdown operation. The tracking method accompanying this invention is implemented in a step-by-step closed loop based on the aforementioned hardware module architecture. It sequentially completes the entire process of system initialization self-check, global communication integration, target data acquisition, tracking range safety determination, hierarchical early warning and control, in-position audible and visual alarm, obstacle density detection, multi-path optimization planning, dynamic flight attitude control, and pre-shutdown processing. Each step is closely connected and linked, relying on AI intelligent algorithms to replace traditional fixed logic control. This enables adaptive judgment, intelligent planning, and dynamic control of UAV dynamic target tracking, significantly improving the accuracy, stability, and safety of dynamic target tracking in complex airspaces. It effectively reduces the probability of UAV flight damage and manual operation costs, and is suitable for fully automated tracking operations in various complex scenarios.

[0024] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An AI-based dynamic target tracking system for unmanned aerial vehicles, characterized in that, include: The system startup initialization module, as the core of the system startup, is used to receive external power-on commands, complete the system-wide hardware self-test, parameter initialization and communication pre-configuration of each sub-module, and output the system startup ready signal; The system main control module receives the ready signal from the system startup initialization module and establishes a full-domain control link. As the core control hub of the system, it is responsible for coordinating the receipt of various sensor data, running AI algorithm logic, issuing control and early warning commands in a hierarchical manner, and coordinating the whole machine tracking operation process. The AI ​​visual positioning and acquisition module establishes a real-time communication link with the system's main control module. It is used to lock onto dynamically tracked targets by relying on the airborne visual recognition unit, collect the target's three-dimensional spatial coordinates and displacement trajectory information in real time, and simultaneously acquire the UAV's real-time flight coordinates, flight attitude, and basic airspace environment data. The tracking range monitoring module pre-stores two levels of height thresholds and two levels of distance thresholds. The two levels of height thresholds include a first height threshold and a second height threshold, and the two levels of distance thresholds include a first distance threshold and a second distance threshold. It is used to calculate the relative height difference and straight-line tracking distance between the UAV and the dynamic target in real time, continuously compare the measured data with the preset thresholds, and determine whether the UAV tracking operation has deviated from the safe and controllable range. The graded early warning module is electrically connected to the tracking range monitoring module and the system main control module. It is used to execute graded early warning logic according to the degree of tracking range exceeding the limit. When the first-level threshold is exceeded, a background pop-up warning is output and the flight attitude is automatically fine-tuned. When the second-level threshold is exceeded, an airborne buzzer warning is activated at the same time, and the over-limit flight displacement of the UAV is restricted. The positioning recognition and audio-visual warning module is connected to the system's main control module. It is used to verify the drone's tracking and positioning status through multiple frames of continuous position data. When it is determined that the drone has reached the preset target tracking point and the tracking task has been completed, it adaptively adjusts the audio-visual intensity to activate the onboard alarm and attracts the attention of people in the surrounding area through audio-visual signals. The all-domain obstacle perception module, with built-in lidar and high-definition camera sensor unit, is used to scan the drone's flight airspace in a grid pattern, collect data on the distribution location, size, and motion status of surrounding obstacles, and calculate the comprehensive distribution density of obstacles by combining obstacle hardness parameters. The AI ​​multi-path intelligent planning module communicates bidirectionally with the all-domain obstacle perception module and the system main control module. When the obstacle density exceeds the preset critical value, it combines multiple parameters such as airspace wind force, geographical topography, obstacle distribution, and flight time to generate multiple feasible navigation paths. It also intelligently filters and scores the multiple paths and pushes the optimal alternative path set to the control terminal for the user to choose from. The flight dynamic control module receives output commands from the system's main control module and the AI ​​multi-path intelligent planning module. It is used to monitor changes in the airspace environment in real time according to the selected navigation path, dynamically fine-tune the UAV's flight attitude, cruise speed and flight path, and avoid environmental damage such as obstacle collisions and airflow impacts. The shutdown preprocessing module is used to receive shutdown termination instructions from the system's main control module and complete pre-shutdown preparations such as backing up tracking operation data, archiving flight parameters, and resetting airborne actuators.

2. The AI-based UAV dynamic target tracking system according to claim 1, characterized in that: The two-level thresholds of the tracking range monitoring module are calibrated through AI big data fitting training. By collecting historical data on the long-term activity radius of the target and the terrain undulation of the tracking area, threshold parameters adapted to the dynamic target movement characteristics are generated. The first height threshold and the first distance threshold are small fluctuation warning thresholds, and the second height threshold and the second distance threshold are safety boundary mandatory alarm thresholds.

3. The AI-based UAV dynamic target tracking system according to claim 1, characterized in that: The tracking range monitoring module calculates the relative height difference and straight-line tracking distance between the UAV and the dynamic target in real time, and combines this with preset two-level height thresholds and two-level distance thresholds to determine whether the tracking operation exceeds the safe and controllable range. The tracking range monitoring module has a built-in precise distance calculation formula, which is used to quantify the straight-line tracking distance between the UAV and the dynamic target in real time, providing data basis for threshold comparison and safe range determination. The formula is as follows: In the formula, Real-time linear tracking distance between the drone and the dynamic target; ( , , () represents the real-time three-dimensional coordinates of the UAV; , , () represents the real-time three-dimensional coordinates of a dynamic target; calculated in real time. The system matches the preset first distance threshold and second distance threshold, and uses the real-time height difference to determine the safe tracking range.

4. The AI-based UAV dynamic target tracking system according to claim 1, characterized in that: The positioning recognition and audio-visual warning module's positioning determination logic involves continuously collecting no less than 5 frames of target position data. Only when the UAV's three-dimensional coordinates continuously fall within the preset point error range and the AI ​​predicts that the target has no significant escape displacement, is the tracking and positioning determined. The audio-visual warning module adaptively adjusts the audio-visual output power according to the ambient light intensity and noise decibels.

5. The AI-based UAV dynamic target tracking system according to claim 1, characterized in that: The global obstacle perception module employs a cubic grid detection mechanism, dividing the airspace surrounding the UAV into several monitoring grids. By fusing LiDAR point cloud data with image semantic recognition data, it calculates the spatial proportion of obstacles within each grid and combines this with a weighted density calculation based on obstacle attributes to accurately determine the airspace congestion level. The overall obstacle density is calculated using a simplified weighted formula, as follows: In the formula, This represents the overall distribution density of obstacles within the grid. This represents the percentage of the space occupied by the obstacle within the detection grid. This is a weighted coefficient for obstacle hardness. The coefficient for hard obstacles is greater than that for soft obstacles, and the coefficient for stationary obstacles is less than that for moving obstacles. This weighted correction improves the accuracy of density detection.

6. The AI-based UAV dynamic target tracking system according to claim 1, characterized in that: The AI ​​multi-path intelligent planning module selects routes based on three core indicators: obstacle density, airspace wind resistance, and flight time. It eliminates inferior routes through a multi-dimensional scoring mechanism and retains 3-5 optimal routes. Each route is also labeled with estimated flight time, risk type, and recommended cruise speed parameters.

7. The AI-based UAV dynamic target tracking system according to claim 1, characterized in that: The flight dynamic control module monitors changes in the airspace environment in real time along the selected navigation path, and makes small adjustments to the UAV's pitch angle, roll angle and flight speed in real time in response to sudden airflow disturbances and temporary obstacle changes.

8. An AI-based method for tracking dynamic targets of unmanned aerial vehicles (UAVs), applied to the AI-based UAV dynamic target tracking system described in any one of claims 1-7, characterized in that, Includes the following steps: Step 1: The system startup initialization module receives the external startup command, completes the whole machine hardware self-test, algorithm parameter initialization and communication configuration of each sub-module, and sends a startup ready signal to the system main control module; Step 2: The system main control module activates the full-domain communication link and sequentially completes the communication connection of the AI ​​visual positioning and acquisition module, the tracking range monitoring module, the full-domain obstacle perception module, and the AI ​​multi-path intelligent planning module. The system then enters standby tracking state. Step 3: The AI ​​visual positioning and acquisition module starts dynamic target recognition, collects target spatial coordinates, displacement trajectory and UAV flight status data in real time, and transmits them synchronously to the system main control module and tracking range monitoring module; Step 4: The tracking range monitoring module calculates the relative height difference and straight-line tracking distance between the UAV and the target in real time, and compares them with the preset two-level height threshold and distance threshold. If the measured data exceeds any level threshold, the graded early warning module is triggered to execute the corresponding early warning action and fine-tune the UAV's flight attitude to return to the safe range. Step 5: The system continuously verifies the tracking point through the positioning recognition and sound and light warning modules. After verifying the tracking position and the target status is stable through multiple frames of data, the system adaptively activates the airborne alarm to provide on-site warning and attract the attention of the surrounding area through sound and light signals. Step Six: During flight, the all-domain obstacle perception module continuously scans the surrounding airspace in a grid pattern, calculates the overall obstacle density with weights, and activates the AI ​​multi-path intelligent planning module when the density exceeds the preset threshold. Step 7: The AI ​​multi-path intelligent planning module combines airspace environment, obstacle distribution, and wind resistance parameters to deduce multiple alternative routes. After multi-dimensional scoring and screening, the routes are pushed to the control terminal, where the user selects the optimal flight path. Step 8: The flight dynamic control module monitors the dynamic changes in the airspace environment in real time according to the path selected by the user, and continuously fine-tunes the flight attitude and route of the UAV to avoid environmental interference such as obstacles and airflow, and protect the safety of the UAV as a whole. Step Nine: After the tracking operation is completed or a shutdown command is received, the shutdown preprocessing module completes preliminary work such as data backup and mechanism reset. After the shutdown status verification module verifies that the shutdown conditions meet the standards, the system closing module completes log archiving, data verification and resource release, and the system stops running.

Images