Cradle head dynamic target anti-interference tracking method fusing multi-sensor data

Abstract

The invention relates to the technical field of holder target tracking methods, in particular to a holder dynamic target anti-interference tracking method fusing multi-sensor data. The method comprises the following steps: firstly, carrying out sub-pixel-level alignment on millimeter wave point cloud, infrared thermal imaging and visible light video data through cascade calibration processing operation, and generating a synchronous data packet containing point cloud position information, a registration temperature matrix, video frame data and a coordinate conversion relation mapping table; based on the data packet, the system adaptively selects an optimal tracking mode through a dynamic decision factor, generates a target space coordinate resolving value through a dominant sensing source type, optimizes trajectory prediction in combination with a multi-source confidence coefficient difference matrix, and drives a holder to execute high-precision anti-interference tracking. According to the method, the technical bottleneck of target binding failure in a high-density shielding environment is broken through, high-efficiency target continuous binding rate and tracking precision are realized in a high-shielding-rate scene, and the problems of tracking interruption and identity confusion caused by insufficient visible light penetrability in a traditional scheme are solved.

Description

Technical Field

[0001] This invention relates to the field of gimbal target tracking methods, and more specifically, to a gimbal dynamic target anti-interference tracking method that integrates multi-sensor data. Background Technology

[0002] In the field of urban public security, the ability of surveillance systems to continuously and stably track suspicious persons and vehicles is directly related to the effectiveness of emergency response. Especially in densely populated areas such as stations and squares, dynamic targets often face the risk of tracking interruption due to the strong obstruction caused by crowds. To address these challenges, existing security systems generally adopt a single visible light visual tracking scheme, using deep learning-based target detection and appearance feature matching mechanisms to attempt to maintain the continuity of target trajectories in complex scenes.

[0003] Although a single visible light visual tracking solution performs reliably in normal scenarios, its identity binding mechanism has fundamental limitations in high-density pedestrian environments with repeated occlusion. Specifically, when the target is occluded by multiple overlapping layers, since visible light cannot penetrate the human body, the system is forced to rely on historical appearance features for target re-identification. Once it encounters interfering personnel with similar clothing, it is very easy to confuse the target's identity, leading to the interruption of tracking association, resulting in gimbal malfunction and target loss. Summary of the Invention

[0004] This invention provides a gimbal-based dynamic target anti-interference tracking method that integrates multi-sensor data. It generates spatiotemporal reference data by synchronously processing millimeter-wave point cloud, infrared thermal imaging, and visible light video data in real time. Based on point cloud distribution characteristics, temperature gradient intensity, and occlusion ratio, it quantifies multi-source confidence levels. Dynamic decision factors enable intelligent arbitration of sensor sources, and the method integrates target spatial coordinate calculation and confidence difference analysis to generate tracking control commands. This solves the problems mentioned in the background art, namely:

[0005] In environments with high-density pedestrians and repeated occlusion, a single visible light tracking system cannot penetrate the human body and is forced to rely on historical appearance features for target re-identification. This is easily affected by similar clothing, leading to target confusion and tracking interruption.

[0006] To achieve the above objectives, the gimbal dynamic target anti-interference tracking method includes the following steps:

[0007] S1. Generate a synchronization data packet containing point cloud location information, registration temperature matrix and video frame data through time alignment and spatial registration processing;

[0008] S2. Perform multi-source feature extraction and map three confidence levels. The steps are as follows:

[0009] S2.1 Generate point cloud discreteness parameters based on point cloud location information and map them to millimeter-wave confidence scores;

[0010] S2.2. Generate temperature gradient intensity values ​​based on the registration temperature matrix and map them to the confidence level of the infrared heat source;

[0011] S2.3. Generate a visible light occlusion ratio coefficient based on the video frame data and map it to an optical reliability coefficient;

[0012] S3. Calculate dynamic decision factors based on three confidence levels and generate a multi-source confidence difference matrix;

[0013] S4. Load the preset threshold and determine the dominant sensor source type based on the dynamic decision factor and the preset threshold;

[0014] S5. Generate target spatial coordinate solution values ​​based on the dominant sensor source type;

[0015] S6. Encode the dominant sensor source type into a dominant sensor source identifier code, and encapsulate it together with the target spatial coordinate solution value and the multi-source confidence difference matrix into a structured control instruction set. Based on the structured control instruction set, predict the dynamic target motion trajectory and calculate the gimbal angle.

[0016] The design concept of the above technical solution stems from the deep integration of the complementary characteristics of multiple sensors: traditional solutions, lacking dynamic quantitative assessment of sensor source reliability, cannot autonomously switch to a more penetrating sensing mode when occlusion density increases dramatically, causing the system to continuously rely on failed visible light data; ignoring the impact of sensor confidence differences on spatial positioning, the appearance of similarly dressed targets will lead to irreversible identity confusion; without establishing a confidence difference early warning mechanism, the accumulation of errors in the prediction stage will lead to a vicious cycle of trajectory drift. This solution, by constructing a sensor reliability quantification system, a dynamic arbitration mechanism, and a multi-dimensional error suppression architecture, forms a triple guarantee mechanism of autonomous activation of penetrating sensors, target thermal feature binding positioning, and blocking of drift propagation sources, systematically overcoming tracking failure and identity confusion in occluded scenarios.

[0017] Based on this, the dynamic arbitration layer performs two-level decision-making:

[0018] Real-time monitoring of sudden changes in feature parameters and triggering compensation verification;

[0019] Output the dominant sensor source command based on the confidence comparison result.

[0020] In another technical solution, the dynamic arbitration layer performs the dominant sensor source type determination as follows:

[0021] When the dynamic decision factor is higher than the first threshold, an infrared dominant command is output.

[0022] When the dynamic decision factor is below the second threshold, an optical dominance command is output.

[0023] Otherwise, output millimeter-wave dominant commands.

[0024] If this technical solution lacks a real-time monitoring and compensation verification mechanism, the system will continuously output distorted confidence scores when there are sudden changes in the temperature field or abnormal jumps in the point cloud, leading to the complete failure of sensor arbitration. If the three-level threshold decision rule is cancelled, the switching of the dominant source in complex occlusion scenarios will lose its quantitative benchmark, reproducing the misjudgment problem of traditional solutions when target speed changes and interference occurs simultaneously. The essential difference between the two lies in the risk defense dimension. The former builds self-healing capabilities for instantaneous anomalies in sensor data, while the latter provides a scenario-adaptive framework for multimodal decision-making: when the ambient temperature changes abruptly, the real-time monitoring mechanism immediately suppresses the attenuation of infrared confidence scores, preserving an error correction window for threshold decision-making; and in scenarios where similar clothing targets are frequently interspersed, the three-level threshold architecture blocks the technical risk of the system mistakenly switching to a failed sensor source by pre-setting optical dominant boundary values.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0026] By nonlinearly coupling millimeter-wave point cloud discreteness with infrared temperature gradient, the physical characteristics of occlusion density are transformed into sensing weight adjustment variables, enabling the system to autonomously strengthen infrared sensing dominance when penetration requirements surge. Simultaneously, an error propagation early warning channel is established using a confidence difference matrix, suppressing the influence weight of low-reliability sensor sources in advance during trajectory prediction, thus blocking the drift accumulation effect caused by open-loop decision-making in traditional solutions at its source. The negative feedback loop formed by these two synergies achieves an order-of-magnitude improvement in target binding reliability in occlusion scenarios. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0028] Figure 2 This is a schematic diagram of the overall process structure of the present invention;

[0029] Figure 3 This is a schematic diagram of the spatiotemporal phase-locked loop module of the present invention;

[0030] Figure 4 This is a schematic diagram of the arbitration decision module of the present invention.

[0031] The meanings of the labels in the diagram are as follows:

[0032] 100. Spatiotemporal phase-locked loop module; 200. Arbitration decision module; 300. Trajectory prediction module; 400. Disturbance rejection control module. Detailed Implementation

[0033] 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.

[0034] Here are some explanations of technical terms:

[0035] RGB color channels refer to a pixel-level encoding system in which visible light sensors achieve full-color reproduction of a target scene by superimposing and mixing the spectral components of the three primary colors: red, green, and blue.

[0036] Currently, visible light tracking systems in high-density occlusion scenarios are prone to target identification confusion, leading to target loss. This invention provides a gimbal-based dynamic target anti-interference tracking method that integrates multi-sensor data. (See [link to relevant documentation]). Figure 1 As shown, it includes the following steps:

[0037] S1. Generate a synchronization data packet containing point cloud location information, registration temperature matrix and video frame data through time alignment and spatial registration processing;

[0038] S2. Perform multi-source feature extraction and map three confidence levels. The steps are as follows:

[0039] S2.1 Generate point cloud discreteness parameters based on point cloud location information and map them to millimeter-wave confidence scores;

[0040] S2.2. Generate temperature gradient intensity values ​​based on the registration temperature matrix and map them to the confidence level of the infrared heat source;

[0041] S2.3. Generate a visible light occlusion ratio coefficient based on the video frame data and map it to an optical reliability coefficient;

[0042] S3. Calculate dynamic decision factors based on three confidence levels and generate a multi-source confidence difference matrix;

[0043] S4. Load the preset threshold and determine the dominant sensor source type based on the dynamic decision factor and the preset threshold;

[0044] S5. Generate target spatial coordinate solution values ​​based on the dominant sensor source type;

[0045] S6. Encode the dominant sensor source type into a dominant sensor source identifier code, and encapsulate it together with the target spatial coordinate solution value and the multi-source confidence difference matrix into a structured control instruction set. Based on the structured control instruction set, predict the dynamic target motion trajectory and calculate the gimbal angle.

[0046] See Figure 2 As shown, the spatiotemporal phase-locked loop module 100, arbitration decision module 200, trajectory prediction module 300, and anti-disturbance control module 400 work together to achieve continuous and accurate tracking of dynamic targets and stable control of the gimbal in complex occlusion environments.

[0047] like Figure 3 As shown, the spatiotemporal phase-locked loop module 100 serves as the data preprocessing center of the system, automatically activating the collaborative acquisition mechanism of the three types of sensors upon startup. This mechanism first synchronously acquires the raw point cloud data from the millimeter-wave radar, the temperature distribution matrix from the infrared thermal imager, and the video frame sequence from the visible light camera. When the environmental monitoring unit detects strong light interference or extreme temperature scenarios, it automatically reduces the sampling frequency of the visible light or infrared sensors; while when the target is moving at high speed, it activates the high-speed scanning mode of the millimeter-wave radar, optimizing resource utilization through dynamic allocation of sensor weights. When the target appears in the monitoring area, the millimeter-wave radar continuously scans to generate raw point cloud data containing three-dimensional position coordinates; the infrared thermal imager simultaneously captures thermal radiation information to form a pixel-level temperature distribution matrix; and the visible light camera outputs a continuous video frame sequence composed of RGB color channels. These three types of raw sensor data are transmitted in parallel via a high-speed bus for subsequent multi-level calibration.

[0048] To address the spatiotemporal misalignment issue of multi-source sensor data, the module performs cascaded calibration processing. The first stage unifies the time reference by generating a synchronization trigger signal using a high-precision clock source and reconstructing the time using sensor delay compensation parameters: raw point cloud data is processed through signal delay interpolation to generate a millimeter-wave point cloud dataset; the pixel-level temperature distribution matrix is ​​re-bound to the starting point of the thermal imaging sampling period; and the video frame sequence is rearranged based on the temporal offset. Different processing strategies are adopted for different motion states: calibration frequency is reduced for static background areas, while sub-millisecond time accuracy calibration is used for moving target areas. The second stage performs spatial registration, mapping infrared and visible light data to a millimeter-wave coordinate system based on pre-calibrated sensor physical position parameters: the temperature distribution matrix is ​​transformed into a registration temperature matrix through nonlinear spatial transformation, and the video frame sequence is transformed through projection to generate synchronized video frames.

[0049] After calibration, the data optimization phase begins. The optimized millimeter-wave point cloud dataset undergoes dynamic target clustering to generate point cloud location information; spatially registered temperature data forms a standard registration temperature matrix; and temporally calibrated video streams generate time-synchronized video frame data. During the encapsulation stage, low-confidence data regions are intelligently filtered: edge points are automatically discarded when the point cloud density is below a threshold, and thermal noise in infrared cold background regions is suppressed. Finally, the spatiotemporal phase-locked loop module 100 combines all optimization results with a coordinate transformation mapping table to form a standardized synchronization data packet, which includes: time-aligned point cloud location information, a registration temperature matrix, video frame data, and a coordinate transformation mapping table. This standardized synchronization data packet is transmitted to subsequent modules, providing a spatially aligned and temporally unified basic dataset for sensor confidence analysis.

[0050] Although the spatiotemporal phase-locked module 100 successfully aligned millimeter-wave point cloud data, infrared temperature distribution matrix, and visible light video frames in the spatiotemporal dimension through a collaborative acquisition mechanism and cascaded calibration processing, establishing a unified data benchmark framework for the system, each sensor source has inherent perception limitations in complex tracking scenarios: visible light sensors cannot penetrate dense crowds, infrared sensors are susceptible to interference from ambient temperature fluctuations, and millimeter-wave sensors lack sufficient accuracy in resolving low-speed target trajectories. The existence of these complementary characteristics necessitates that the system must evaluate the reliability of each sensor source in real time. To address this, we introduce the arbitration decision module 200. This module directly receives the point cloud position information, registration temperature matrix, video frame data, and coordinate transformation relationship mapping table processed by the spatiotemporal phase-locked module 100, and determines the optimal sensor source strategy in real time through a dynamic confidence evaluation mechanism.

[0051] like Figure 4 As shown, the arbitration decision module 200 receives point cloud location information, registration temperature matrix, video frame data, and coordinate transformation relationship mapping table transmitted by the spatiotemporal phase-locked module 100. The module first performs parallel analysis of multi-source features: it calculates the point cloud distribution characteristics of the target area using the three-dimensional coordinate set in the point cloud location information, generating point cloud discreteness parameters characterizing the degree of spatial discreteness; it simultaneously extracts the thermal radiation data of the target-enclosed area from the registration temperature matrix, and calculates the core temperature gradient intensity value using a spatial gradient operator; simultaneously, it performs pixel-level visibility analysis on the video frame data, statistically analyzes the proportion of occluded pixels within the target area, and outputs the visible light occlusion ratio coefficient. During this process, the coordinate transformation relationship mapping table is continuously invoked to ensure that all feature parameters are calculated under a unified three-dimensional spatial reference.

[0052] The core innovation of the arbitration decision module 200 lies in establishing a confidence dynamic game engine, which includes a hierarchical decision-making mechanism. In the basic evaluation layer, a feature quantization model is used: a negative exponential transformation is applied to the point cloud discreteness parameters, mapping them to a millimeter-wave confidence score in the [0,1] interval, where a larger discreteness results in a lower score; the temperature gradient intensity value is input into an improved response function, outputting the infrared heat source confidence level, with a higher level for steeper temperature field changes; the optical confidence coefficient is derived based on the visible light occlusion ratio coefficient, showing a linear negative correlation between the occlusion ratio and confidence. These confidence parameters constitute a three-dimensional decision vector, and the core formula for intelligent arbitration through dynamic decision factors is:

[0053] ;

[0054] In the formula, Indicates dynamic decision factors;

[0055] Indicates infrared confidence level;

[0056] It is the natural constant (approximately 2.71828);

[0057] Indicates the occlusion response coefficient;

[0058] This represents the point cloud discreteness parameter;

[0059] Indicates the confidence level of millimeter waves;

[0060] Indicates optical confidence level;

[0061] It is a very small constant (0.01).

[0062] The resulting dynamic decision factors and three-dimensional confidence vector together form the basis of system decision-making, realizing a closed-loop logic from multi-source data evaluation to tracking strategy formulation.

[0063] A two-level decision-making rule is set in the dynamic arbitration layer: The first level performs real-time status diagnosis, continuously monitors the time-varying differential characteristics of the temperature gradient intensity value, and activates the thermal interference compensation mechanism when a gradient value mutation exceeds 30%, automatically reducing the confidence weight of the infrared heat source; it also tracks the continuous frame fluctuations of the point cloud discreteness parameter, and triggers millimeter-wave trajectory authenticity verification when a non-continuous jump occurs. The second level implements the dominant source decision tree: if the infrared heat source confidence level is consistently higher than the optical confidence coefficient by more than 0.4 and the millimeter-wave confidence score is lower than the dynamic threshold (which is adaptively adjusted according to the ambient temperature), then an infrared dominant command is output; if the point cloud discreteness parameter is lower than the density tolerance value for three consecutive frames and is accompanied by abnormal attenuation of the temperature gradient, then the millimeter-wave tracking mode is activated and the visible light data stream is suspended.

[0064] The entire decision-making process incorporates a closed-loop feedback mechanism: after the arbitration result is output to the trajectory prediction module 300, the trajectory fitting residual data is received in real time to optimize the confidence model parameters in reverse; the infrared-dominated command scenario is further verified by heating radiation distribution, and if the temperature gradient around the target does not conform to the distribution law of human body thermal field, the decision rollback is triggered.

[0065] After completing the two-level decision-making process of the dynamic arbitration layer, the arbitration decision module 200 immediately generates three core outputs based on the final decision result. First, the module determines the current dominant sensor type according to the decision rules, encodes it into a machine-readable identifier, distinguishes between millimeter-wave dominant, infrared dominant, and optical dominant control modes, and generates a dominant sensor source identifier. Then, it initiates the target spatial coordinate calculation process to generate the target spatial coordinate solution value: when in infrared dominant mode, it fuses the hotspot coordinates in the registration temperature matrix with the point cloud height value to generate the three-dimensional position; in millimeter-wave dominant mode, it directly outputs the spatial coordinates of the point cloud cluster center; if optical dominant, it combines the video detection box center coordinates with the depth estimate to construct spatial positioning. Simultaneously, the module calculates the pairwise differences between millimeter-wave confidence, infrared confidence, and optical confidence, forming a multi-source confidence difference matrix describing the reliability differences among multiple sensor sources.

[0066] The entire processing incorporates a closed-loop feedback mechanism for real-time verification. Ultimately, the generated dominant sensor source identifier code, target spatial coordinate solution value, and multi-source confidence difference matrix are encapsulated into a structured control instruction set, which is transmitted to the trajectory prediction module 300 via a high-speed interface. The target spatial coordinate solution value will serve as the starting point for trajectory modeling, the dominant sensor source identifier code will determine the selection of prediction algorithm branches, and the multi-source confidence difference matrix will be used to compensate for prediction uncertainties.

[0067] The trajectory prediction module 300 receives a structured control instruction set transmitted by the arbitration decision module 200, which includes the dominant sensor source identifier code, the target spatial coordinate solution value, and the multi-source confidence difference matrix. The module first parses the dominant sensor source identifier code to determine the prediction algorithm path: when the identifier code indicates millimeter wave dominance, the spatial coordinates of the point cloud cluster center in the target spatial coordinate solution value are directly used as the trajectory modeling reference point; if the identifier is infrared dominance, the temperature matrix hotspot coordinates and point cloud height values ​​within the instruction set are fused to construct the 3D tracking starting point; when the identifier is optical dominance, the spatial positioning reference is established by combining the video detection box center coordinates and depth estimation values. During this process, the multi-source confidence difference matrix is ​​loaded in real time. By analyzing the pairwise differences in confidence values ​​between millimeter wave, infrared, and optical sources, the weighting coefficients of each sensor source in the prediction model are dynamically adjusted. Sensor sources with confidence differences exceeding a threshold will have their weighting factors reduced.

[0068] Based on the determined starting coordinates and weighting parameters, the module initiates an adaptive trajectory prediction algorithm. This algorithm includes a motion state prediction mechanism: continuously detecting the acceleration changes of coordinate displacement within five consecutive frames, and automatically switching to millimeter-wave assisted correction mode when the instantaneous acceleration exceeds the limits of human motion; simultaneously, it monitors the angular velocity of the target direction and introduces inertial compensation parameters in sharp turning scenarios. In the trajectory calculation phase, the module combines historical trajectory point sets with current motion vector features to generate a predicted trajectory coordinate sequence of the target in three-dimensional space through Bézier curve fitting. This sequence includes the displacement path and direction change trend within the next three frames.

[0069] The entire prediction process incorporates a quality verification step: the trajectory fitting results are back-projected onto the coordinate systems of each sensor source, and anomaly alarm items in the multi-source confidence difference matrix are compared; when the difference between infrared and optical confidence exceeds the safety tolerance, a trajectory smoothing filter is triggered to suppress abrupt noise. Finally, the module generates a target trajectory prediction package, which includes a timestamp-aligned prediction coordinate sequence, motion state vectors (velocity / acceleration / direction angle), and a trajectory reliability score. This data package is transmitted in its entirety to the anti-disturbance control module 400 via a real-time communication interface, providing it with a spatial trajectory reference for calculating the gimbal rotation angle.

[0070] The anti-interference control module 400 receives the target trajectory prediction packet transmitted by the trajectory prediction module 300, which includes a timestamp-aligned predicted coordinate sequence, motion state vector (velocity / acceleration / direction angle), and trajectory reliability score. The module first parses the predicted coordinate sequence to extract the target's displacement path for the next three frames, generating a gimbal aiming angle reference value. Simultaneously, it analyzes the acceleration parameters in the motion state vector. When a sudden acceleration change exceeds a preset threshold, a rapid compensation mechanism is activated to generate an angle compensation offset. The control response gain coefficient is dynamically adjusted based on the trajectory reliability score; the lower the score, the smaller the gain coefficient to suppress trajectory jitter. During this process, the module synchronously acquires sensor source reliability difference data stored in the multi-source confidence difference matrix, applying weighted suppression factors to sensor sources with confidence differences exceeding the safety tolerance to ensure that low-reliability sensor sources do not affect the final control quality.

[0071] Based on the target motion characteristics provided by trajectory prediction, the anti-disturbance control module 400 performs intelligent fusion calculations of gimbal control commands. The module first establishes a three-dimensional spatial motion mapping mechanism: dynamically combining the predicted angle benchmark of the target's future position with the actual motion compensation amount to generate precise horizontal and vertical rotation commands. This process employs an adaptive control strategy, driving the gimbal to quickly respond to target displacement in high-confidence scenarios and automatically switching to a smooth tracking mode in low-reliability states, effectively suppressing sudden jitter. Addressing the confidence differences between sensor sources, a unique dynamic weight allocation algorithm automatically reduces the influence weight of abnormal data sources. The entire system incorporates dual safety safeguards: actively offsetting equipment resonance interference through real-time environmental vibration monitoring and simultaneously verifying the compliance of the motion trajectory, ensuring that control commands always conform to the actual displacement trend of the target.

[0072] The final generated anti-disturbance control command includes three core technical parameters, which are used to adjust the gimbal movement in real time through a high-precision servo system: the three-dimensional rotation angle accurately calculates the target's azimuth offset in space, converting the trajectory prediction coordinates into the gimbal's horizontal yaw angle and vertical pitch angle; the dynamic velocity curve is based on the target motion state analysis results, dynamically generating a profile of the gimbal's rotation rate change to ensure that the mechanical response speed and target displacement remain smoothly matched; the environmental interference suppression parameter continuously monitors the equipment's vibration spectrum characteristics, generates a phase reversal signal and injects it into the drive circuit to cancel the angle drift caused by airflow disturbance and mechanical resonance in real time.

[0073] These three types of parameters work together to construct an adaptive closed loop: the rotation angle provides a spatial positioning reference, the velocity curve couples the motion trajectory, and the interference suppression parameter maintains execution stability. When the system detects a sudden change in target motion caused by high-density occlusion, the dynamic velocity curve immediately adjusts the gimbal acceleration, while the interference suppression module enhances vibration resistance. When the optical sensor restores target visibility, the three-dimensional rotation angle synchronously optimizes positioning accuracy to achieve seamless tracking. The entire control process is supported by multi-sensor data collaboration, ultimately enabling the gimbal to maintain spatial locking on dynamic targets even under interference in multiple scenarios, forming a technical closed loop from perception and decision-making to control execution.

[0074] This invention converts the predicted spatial position into a gimbal control angle in real time, dynamically adjusts the rotation speed curve based on the target's motion state, and actively compensates for environmental vibration interference. This technology chain achieves stable tracking through a three-order collaborative mechanism: first, it establishes a precise spatial coordinate mapping relationship to ensure lossless conversion from the target position to the gimbal angle; second, it generates a smooth velocity control curve through a motion state adaptive algorithm to maintain motion coupling between the gimbal and the target; and finally, it injects a real-time vibration suppression signal to eliminate the impact of mechanical drift on positioning accuracy. This successfully overcomes the target identity confusion problem caused by high-density occlusion, maintaining continuous and stable tracking of dynamically moving objects even in scenarios with repeated pedestrian occlusion, ultimately achieving precise tracking control at the spatial locking level.

[0075] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A gimbal-based dynamic target anti-interference tracking method that integrates multi-sensor data, characterized in that, Includes the following steps: S1. Generate a synchronization data packet containing point cloud location information, registration temperature matrix and video frame data through time alignment and spatial registration processing; S2. Perform multi-source feature extraction and map three confidence levels. The steps are as follows: S2.1 Generate point cloud discreteness parameters based on point cloud location information and map them to millimeter-wave confidence scores; S2.

2. Generate temperature gradient intensity values ​​based on the registration temperature matrix and map them to the confidence level of the infrared heat source; S2.

3. Generate a visible light occlusion ratio coefficient based on the video frame data and map it to an optical reliability coefficient; S3. Calculate dynamic decision factors based on three confidence levels and generate a multi-source confidence difference matrix; S4. Load the preset threshold and determine the dominant sensor source type based on the dynamic decision factor and the preset threshold; S5. Generate target spatial coordinate solution values ​​based on the dominant sensor source type; S6. Encode the dominant sensor source type into a dominant sensor source identifier code, and encapsulate it together with the target spatial coordinate solution value and the multi-source confidence difference matrix into a structured control instruction set. Based on the structured control instruction set, predict the dynamic target motion trajectory and calculate the gimbal angle.

2. The gimbal dynamic target anti-interference tracking method based on multi-sensor data fusion according to claim 1, characterized in that: The sensing weights are dynamically allocated through the spatiotemporal phase-locked module (100), and millimeter-wave point cloud data, infrared thermal imaging data and visible light video data are collected simultaneously.

3. The gimbal dynamic target anti-interference tracking method based on multi-sensor data as described in claim 2, characterized in that: The time-space phase-locked module (100) performs timestamp alignment and spatial coordinate system registration on millimeter-wave point cloud data, infrared thermal imaging data and visible light video data to generate the synchronization data packet.

4. The gimbal dynamic target anti-interference tracking method based on multi-sensor data as described in claim 1, characterized in that: A confidence dynamic game engine is established through the arbitration decision module (200), which includes a hierarchical decision mechanism, comprising a basic evaluation layer and a dynamic arbitration layer.

5. The gimbal dynamic target anti-interference tracking method based on multi-sensor data according to claim 4, characterized in that: The dynamic arbitration layer performs two-level decision-making: Real-time monitoring of sudden changes in feature parameters and triggering compensation verification; Output the dominant sensor source command based on the confidence comparison result.

6. The gimbal dynamic target anti-interference tracking method based on multi-sensor data according to claim 5, characterized in that: The dynamic arbitration layer determines the dominant sensor source type as follows: When the dynamic decision factor is higher than the first threshold, an infrared dominant command is output. When the dynamic decision factor is below the second threshold, an optical dominance command is output. Otherwise, output millimeter-wave dominant commands.

7. The gimbal dynamic target anti-interference tracking method based on multi-sensor data according to claim 1, characterized in that: The trajectory prediction module (300) selects a trajectory prediction algorithm based on the dominant sensor source identifier code and loads a multi-source confidence difference matrix to dynamically adjust the sensor source weights, thereby generating a target trajectory prediction package.

8. The gimbal dynamic target anti-interference tracking method based on multi-sensor data according to claim 7, characterized in that: The target trajectory prediction package includes a timestamp-aligned predicted coordinate sequence, motion state vector, and trajectory reliability score, and is transmitted to the disturbance rejection control module (400).

9. The gimbal dynamic target anti-interference tracking method according to claim 8, which integrates multi-sensor data, is characterized in that: The anti-disturbance control module (400) converts the predicted coordinate sequence into a gimbal aiming angle reference value, generates an angle compensation offset by combining the motion state vector, and adjusts the control gain coefficient based on the trajectory reliability score.

10. The gimbal dynamic target anti-interference tracking method based on multi-sensor data according to claim 9, characterized in that: The anti-disturbance control module (400) integrates the pre-aiming angle reference value, angle compensation offset and gain coefficient to generate an anti-disturbance control signal, which drives the gimbal to perform dynamic target tracking.

Images