An unmanned aerial vehicle-based mobile target positioning and tracking method

Abstract

This disclosure provides a mobile target localization and tracking method based on unmanned aerial vehicles (UAVs). The method includes: acquiring container tunnel structure data and performing spatial topology modeling based on the container tunnel structure data to construct a target passage space constraint model; extracting spatial projection points of the tracked target within the tunnel spatial constraint zone from the target passage space constraint model, and generating the target tunnel process state based on the spatial projection points; identifying occlusion interval partitions through the target tunnel process state, and determining the target continuity reliability based on the occlusion interval partitions; reconstructing the UAV escort path based on the target continuity reliability, and generating target tracking and positioning results by executing the reconstructed UAV escort path. This disclosure addresses the technical problem of periodic, aligned occlusion in port container tunnels leading to target substitution and escort path errors, achieving target uniqueness maintenance and stable continuous tracking in narrow passage environments.

Description

Technical Field

[0001] This disclosure pertains to the field of unmanned aerial vehicle (UAV) technology, and more specifically, relates to a method for locating and tracking mobile targets based on UAVs. Background Technology

[0002] In scenarios such as container terminals, logistics yards, and automated warehouse outdoor areas, moving targets traveling along container aisles repeatedly enter periodic obscuring zones formed by neatly arranged containers. This obscuring is not a one-off, short-term obscuring, but rather exhibits a regular alternating pattern of appearance, disappearance, and reappearance. Due to the narrow and elongated dimensions of the aisles, the similar height of the side walls, and the partially open tops, the target is briefly visible each time it passes through an aisle opening, and immediately becomes invisible again after entering a continuous section of containers.

[0003] Currently, traditional positioning and tracking solutions can maintain short-term tracking in general scenarios by relying on position estimation and trajectory extrapolation. However, in the aforementioned periodic, truncated environments, drones are prone to continuously mistaking multiple reappearing nearby targets at openings for the same object, or forming incorrect flight paths over narrow passageways. The difficulty lies in the fact that the truncated structures are caused by the fixed layout of the engineering site and cannot be completely eliminated simply by adding sensing devices. Furthermore, the box-shaped passageways compress the target's degrees of freedom of movement into strongly channelized movement, causing multiple targets to appear in close proximity, making it difficult for existing technical solutions to maintain a unique and stable target orientation in port passageway environments. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to overcome the aforementioned deficiencies and propose a mobile target localization and tracking method based on unmanned aerial vehicles (UAVs).

[0005] The present invention adopts the following technical solution.

[0006] The first aspect of this invention discloses a method for locating and tracking a moving target based on an unmanned aerial vehicle (UAV), the method comprising:

[0007] Obtain container tunnel structure data and perform spatial topology modeling based on the container tunnel structure data to construct a target passage space constraint model;

[0008] Extract the spatial projection points of the tracking target within the lane spatial constraint zone from the target passage space constraint model, and generate the target lane process state based on the spatial projection points;

[0009] The occlusion interval partitions are identified by the target roadway process status, and the target continuity confidence is determined based on the occlusion interval partitions;

[0010] Based on the target's continuity credibility, the drone's flight path is reconstructed, and the target tracking and positioning result is generated by executing the reconstructed drone flight path.

[0011] Furthermore, the step of acquiring container tunnel structure data and performing spatial topology modeling based on the container tunnel structure data to construct a target passage space constraint model includes:

[0012] The drone conducts a cruise scan of the container tunnel area to obtain cruise scan data, and uses a laser rangefinder to determine the edge points of the containers on both sides of the tunnel, so as to generate a tunnel boundary point sequence based on the cruise scan data and the container edge points.

[0013] The roadway boundary point sequence is subjected to sliding window smoothing fitting and roadway centerline discrete sampling to obtain the roadway boundary curve and roadway centerline discrete sequence.

[0014] Furthermore, the step of acquiring container tunnel structure data and performing spatial topology modeling based on the container tunnel structure data to construct a target passage space constraint model further includes:

[0015] Multiple sampling sections are selected in the longitudinal direction of the roadway, and the boundary distance between the two sides of the roadway in the sampling section is calculated according to the roadway boundary curve and the discrete sequence of the roadway centerline. The boundary distance is then corrected for fluctuations to obtain the effective passage width of the roadway.

[0016] The entity width and swing margin of the tracked target and the positioning safety boundary of the UAV are obtained. The effective passage width of the alley is used as a constraint. Based on the entity width and swing margin of the tracked target and the positioning safety boundary of the UAV, the alley space constraint zone is constructed.

[0017] Based on the discrete sequence of the lane space constraint zone and the lane centerline, the target passage space constraint model is generated.

[0018] Furthermore, the step of extracting the spatial projection points of the tracking target within the roadway spatial constraint zone from the target passage space constraint model, and generating the target roadway progress state based on the spatial projection points, includes:

[0019] The initial spatial location of the target within the lane spatial constraint zone is obtained, and multiple candidate spatial projection points corresponding to the initial spatial location are selected from the target passage spatial constraint model. The distance between the candidate spatial projection points and the initial spatial location does not exceed a set threshold.

[0020] Calculate the geometric error between the candidate spatial projection point and the initial spatial position point at the same time, and call the objective function to select the spatial projection point with the smallest function value at the corresponding time based on the geometric error.

[0021] Furthermore, the step of extracting the spatial projection points of the tracking target within the roadway spatial constraint zone from the target passage space constraint model, and generating the target roadway progress state based on the spatial projection points, further includes:

[0022] The spatial projection points are sorted in chronological order to construct a spatial projection point sequence, and the segment length, inter-segment turning point, and local jump variable between adjacent spatial projection points in the spatial projection point sequence are calculated.

[0023] Abnormal segments are eliminated based on the segment length, inter-segment transitions, and local jump variables to obtain the target travel path of the tracking target within the spatial constraint zone of the roadway, and the spatial projection points in the target travel path are converted into roadway arc length positions.

[0024] The target roadway process state is obtained by integrating the spatial projection point and the corresponding roadway arc length position at a unified moment. The roadway arc length position is the cumulative arc length between the spatial projection point and the corresponding roadway centerline reference point in the target passage space constraint model.

[0025] Furthermore, the step of identifying occlusion interval partitions through the target roadway process state and determining the target continuity confidence based on the occlusion interval partitions includes:

[0026] Calculate the change in arc length between adjacent arc length positions in the roadway, determine the arc length discontinuity difference between different segments based on the change in arc length, and convert the arc length discontinuity difference into a candidate occlusion trigger value.

[0027] An initial trigger threshold is set based on the average distribution of the candidate occlusion trigger quantity across multiple sampling intervals, and sampling intervals where the candidate occlusion trigger quantity exceeds the initial trigger threshold are marked as candidate occlusion trigger points.

[0028] The candidate occlusion trigger points are time-series aggregated to merge multiple candidate occlusion trigger points at adjacent times into the occlusion interval partition, and the occlusion intensity corresponding to the occlusion interval partition is calculated.

[0029] Furthermore, the step of identifying occlusion interval partitions through the target roadway process state and determining the target continuity confidence based on the occlusion interval partitions also includes:

[0030] Obtain the duration corresponding to each occlusion interval partition, and filter the occlusion interval partitions according to the duration and occlusion intensity to retain occlusion interval partitions whose duration and occlusion intensity meet the set threshold range;

[0031] Valid occlusion intervals are selected from the occlusion interval partitions, and the arc length start point and arc length end point corresponding to the valid occlusion intervals are determined, so as to calculate the predicted arc length position of the tracking target at the end point of the valid occlusion interval based on the valid occlusion intervals and the corresponding arc length start point and arc length end point.

[0032] The actual arc length position of the tracked target from the arc length start point to the arc length end point is obtained, and the arc length error between the predicted arc length position and the actual arc length position is calculated. The continuity confidence of the target is determined based on the arc length error, and the arc length error is inversely proportional to the continuity confidence of the target.

[0033] Furthermore, the process of reconstructing the UAV escort path based on the target continuity confidence, and generating target tracking and positioning results by executing the reconstructed UAV escort path, includes:

[0034] The optimal target continuation chain is determined based on the target continuation credibility, and a drone escort reference path is constructed based on the optimal target continuation chain;

[0035] The drone escort path is converged and updated using the drone escort reference path to obtain the reconstructed drone escort path. Based on the reconstructed drone escort path, the optimal target continuation chain and the drone escort path at each moment during the escort process are associated, accumulated, and stored to obtain the target tracking and positioning result.

[0036] A second aspect of the present invention discloses a mobile target localization and tracking device based on an unmanned aerial vehicle (UAV), used to implement the mobile target localization and tracking method based on an UAV as described in any one of the first aspects, the device comprising:

[0037] The constraint model construction module is used to acquire container tunnel structure data and perform spatial topology modeling based on the container tunnel structure data to construct a target passage space constraint model.

[0038] The process state generation module is used to extract the spatial projection points of the tracking target within the lane spatial constraint zone from the target access space constraint model, and generate the target lane process state based on the spatial projection points.

[0039] The target continuity identification module is used to identify the occlusion interval partitions through the target roadway progress status, and determine the target continuity confidence based on the occlusion interval partitions;

[0040] The tracking and positioning output module is used to reconstruct the drone's flight path based on the target's continuity reliability, and generate the target tracking and positioning result by executing the reconstructed drone flight path.

[0041] A third aspect of the present invention discloses a terminal, including a processor and a storage medium;

[0042] The storage medium is used to store instructions;

[0043] The processor is configured to operate according to the instructions to perform the steps of the method described in the first aspect.

[0044] A fourth aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0045] The beneficial effects of the present invention are as follows: Compared with the prior art, the present invention has the following advantages:

[0046] (1) This invention acquires container tunnel structure data and performs spatial topology modeling based on the container tunnel structure data to construct a target passage space constraint model. Then, it extracts the spatial projection points of the tracking target within the tunnel space constraint zone from the target passage space constraint model and generates the target tunnel process state based on the spatial projection points. Subsequently, it identifies the occlusion interval partitions through the target tunnel process state and determines the target continuation credibility based on the occlusion interval partitions. Addressing the technical problem of periodic full-row occlusion in port container tunnels causing target substitute continuation and flight path errors during the appearance, disappearance, and reappearance of targets, this invention identifies whether the tracking target reappears after occlusion by constructing occlusion intervals and target continuation credibility, thereby achieving target uniqueness maintenance and stable continuous tracking in narrow passage environments.

[0047] (2) Based on the reliability of target continuity, the present invention reconstructs the drone's flight path and generates the final target tracking and positioning result by executing the reconstructed drone flight path. This enables the drone to flexibly optimize the flight path according to whether the target being tracked is obscured during the tracking process, thereby improving the flexibility of drone flight tracking. Combined with the verification of target continuity in the obscured area, the accuracy and continuity of tracking a unique target in a periodically occluded environment are further improved. Attached Figure Description

[0048] Figure 1 This is a flowchart illustrating a mobile target localization and tracking method based on an unmanned aerial vehicle (UAV) provided by the present invention.

[0049] Figure 2 This is a schematic diagram of the structure of a mobile target positioning and tracking device based on an unmanned aerial vehicle (UAV) provided by the present invention. Detailed Implementation

[0050] The present application will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, and should not be construed as limiting the scope of protection of the present application.

[0051] like Figure 1 As shown, in one embodiment, a mobile target localization and tracking method based on an unmanned aerial vehicle (UAV) includes the following steps:

[0052] Step S110: Obtain container tunnel structure data and perform spatial topology modeling based on the container tunnel structure data to construct a target passage space constraint model.

[0053] In some embodiments, the mobile target localization and tracking method based on a drone provided by the present invention includes the following steps in step S110:

[0054] Step S111: A drone is used to perform a cruise scan in the container tunnel area to obtain cruise scan data. A laser rangefinder is used to determine the edge points of the containers on both sides of the tunnel, so as to generate a tunnel boundary point sequence based on the cruise scan data and the container edge points.

[0055] Understandably, the lane boundary point sequence is used to characterize the spatial distribution of the boundaries on both sides of the container lane, providing a basis for lane boundary curve fitting, lane centerline discrete sampling, and calculation of the effective passage width of the lane. Specifically, the edge points of the containers on both sides can be extracted first using cruise scan data, then sorted according to the lane extension direction, and categorized by left and right sides and weighted by cross-section to form the lane boundary point sequence.

[0056] Step S112: Perform sliding window smoothing fitting and discrete sampling of the roadway centerline on the roadway boundary point sequence to obtain the roadway boundary curve and the discrete sequence of the roadway centerline.

[0057] In some embodiments, the mobile target localization and tracking method based on a drone provided by the present invention further includes the following steps in step S110:

[0058] Step S113: Select multiple sampling sections in the longitudinal direction of the roadway, and calculate the boundary distance between the two sides of the roadway in the sampling section according to the roadway boundary curve and the discrete sequence of the roadway centerline, so as to correct the fluctuation of the boundary distance and obtain the effective passage width of the roadway.

[0059] Step S114: Obtain the entity width and swing margin of the tracked target and the positioning safety boundary of the UAV. Using the effective passage width of the alley as a constraint, construct the alley space constraint zone based on the entity width and swing margin of the tracked target and the positioning safety boundary of the UAV.

[0060] Step S115: Generate the target passage space constraint model based on the discrete sequence of the lane space constraint zone and the lane centerline.

[0061] In a specific embodiment, the present invention provides a mobile target localization and tracking method based on an unmanned aerial vehicle (UAV), comprising steps 1 to 4:

[0062] Step 1: Perform lane space topology modeling and target passage zone constraint construction.

[0063] Includes the following sub-steps:

[0064] Sub-step 1.1: Construct the left boundary point sequence and the right boundary point sequence sorted according to the direction of the roadway extension.

[0065] Specifically, firstly, a drone performs a low-speed cruise scan along the top of the container aisle, using laser ranging or structured light ranging equipment to acquire edge points on the outer walls of the containers on both sides of the aisle. Due to the distinctly long and narrow structure of port container aisles, all sampling points are first sorted longitudinally based on the main extension direction of the aisle. Then, according to the positive and negative relationship of the lateral position of the midpoint of each longitudinal section, the sampling points are divided into left and right boundary candidate points. Subsequently, a weighted average is performed on multiple candidate points near the same longitudinal position; that is, the weighted average is used to suppress the influence of isolated noise points on the boundary position, making the representative boundary points closer to the actual container edges, thus obtaining representative points for the left and right sides of the cross-section. These are then connected sequentially to form the left and right boundary point sequences.

[0066] Sub-step 1.2: Construct a continuous left boundary curve, right boundary curve, and discrete sampling sequence of the roadway centerline.

[0067] Specifically, considering that although the container stacking is generally regular, there may be slight misalignments, protruding corner pieces, or protruding door locks in some areas, the original discrete points cannot be directly used as the final boundaries. Therefore, it is necessary to perform sliding window smoothing fitting on the left and right boundary point sequences separately. During fitting, the vertical index is used as the independent variable, and the horizontal and vertical coordinates are locally smoothed respectively. Then, the midpoint of the left and right boundary curves is taken at the same vertical position to obtain the discrete sampling points of the tunnel centerline, which are then connected to form the tunnel centerline function.

[0068] In this embodiment, the expression for the horizontal coordinate of the centerline is:

[0069]

[0070] In the formula, Indicates the first The lateral coordinates of each centerline sampling point are in meters. , These represent the lateral coordinates of the left and right boundary points at the same longitudinal section, in meters; , These represent the height coordinates of the left and right boundary points, respectively, in meters; This represents the height correction reference value, in meters, ranging from 2.0 to 6.0.

[0071] It should be noted that if the center line is obtained by simply dividing the center, when the height difference between the stacked boxes is large, the passageway constructed later will be offset to the side of the higher stack or the side of the lower stack. This will cause the target to be misjudged as deviating from the passage when it is clearly located in the actual passage. By continuous fitting and height offset correction, the subsequent passageway model can be made closer to the actual passable area.

[0072] Sub-step 1.3: Calculate the average width of the roadway, the width fluctuation correction value, and the effective passage width of the roadway.

[0073] Specifically, multiple sampling sections are selected along the longitudinal direction of the tunnel, and the lateral clearance between the left and right boundaries is calculated for each section. Fluctuation corrections are then applied to the width based on local misalignment of containers, bulging corners, and tunnel bends. Since port tunnels are not perfectly regular corridors, simply taking the arithmetic mean of all section widths may incorrectly diffuse a few abnormally wide or narrow points into the overall model. Therefore, a dispersion suppression term needs to be added to the average width to obtain the effective passage width.

[0074] The expression for the effective passage width is:

[0075]

[0076] In the formula, Indicates the effective passage width of the alleyway, in meters; Indicates the first The net distance between the left and right boundaries of the sampling section, in meters, is derived from the first... The absolute value of the lateral difference between the left and right boundary points of each section is obtained; This represents the average net distance across all sampling sections, in meters. This represents the total number of sampling sections, with a value range of 20-200. This represents the width fluctuation suppression coefficient, a dimensionless quantity, with a value ranging from 0.3 to 0.8. The value is set according to the neatness of the container placement on the site; 0.3-0.5 is used when the neatness is high, and 0.5-0.8 is used when there is significant misalignment. This formula is used to correct what appears to be a sufficiently wide width to a width that allows stable passage in most locations, thereby preventing the passageway from being built too wide due to local container misalignment.

[0077] Sub-step 1.4: Construct the set of lane space constraint zones and the target passage space constraint model.

[0078] Specifically, a spatial constraint zone is constructed using the centerline of the tunnel as the framework and half the effective passage width of the tunnel as the lateral constraint radius. This also considers the target entity width, target swing margin, and positioning safety boundary in the UAV tracking scenario. That is, for any spatial point to be judged, the lateral distance to the nearest sampling point on the centerline is first calculated and then compared with the constraint radius. When this distance is less than or equal to the constraint radius, the spatial point is determined to be within the allowable passage zone. For partially open areas at the top of the tunnel, the entire open space above is not directly included in the constraint zone. Instead, it is still projected downwards along the centerline to form the main passage framework, ensuring that subsequent target movement is always confined to the spatial zone corresponding to the box-shaped passage, rather than spreading to adjacent tunnels or intersecting work areas.

[0079] In this embodiment, the expression for the spatial constraint determination radius is:

[0080]

[0081] In the formula, This indicates the spatial constraint radius of the tunnel, in meters. This represents the effective passage width of the roadway obtained in sub-step 1.3, in meters; This represents the equivalent lateral width of the target entity, in meters. For example, pedestrians are represented by 0.4-0.8 meters, forklifts by 1.2-2.5 meters, and transport vehicles by 1.8-3.0 meters. This represents the target width reduction factor, which is a dimensionless quantity and ranges from 0.4 to 0.7. This represents the safety setback boundary, expressed in meters, with a value ranging from 0.1 to 0.6. This formula subtracts the target's own occupancy and the safety boundary from the geometrically passable width to obtain the effective constraint radius truly used for tracking and discrimination.

[0082] Step S120: Extract the spatial projection points of the tracking target within the roadway spatial constraint zone from the target passage space constraint model, and generate the target roadway process status based on the spatial projection points.

[0083] In some embodiments, the mobile target localization and tracking method based on a drone provided by the present invention includes the following steps in step S120:

[0084] Step S121: Obtain the initial spatial location of the target within the lane spatial constraint zone, and select multiple candidate spatial projection points corresponding to the initial spatial location from the target passage spatial constraint model. The distance between the candidate spatial projection points and the initial spatial location does not exceed a set threshold.

[0085] Among them, candidate spatial projection points are used to provide multiple reasonable mapping positions for the initial spatial location points within the lane spatial constraint zone, so as to avoid the target passage path being mistakenly compressed or mistakenly attracted by direct single-point projection.

[0086] Step S122: Calculate the geometric error between the candidate spatial projection point and the initial spatial position point at the same time, and call the objective function to select the spatial projection point with the smallest function value at the corresponding time based on the geometric error.

[0087] Geometric error is used to measure the degree of spatial fit between the candidate spatial projection point and the initial spatial position point. During calculation, the lateral coordinate deviation, longitudinal coordinate deviation, height coordinate deviation, and lateral deviation relative to the central skeleton of the lane spatial constraint zone are comprehensively compared to obtain a comprehensive evaluation result for a single candidate spatial projection point. In some embodiments, the UAV-based moving target localization and tracking method provided by the present invention further includes the following steps in step S120:

[0088] Step S123: Sort the spatial projection points in chronological order to construct a spatial projection point sequence, and calculate the segment length, inter-segment transition, and local jump variables between adjacent spatial projection points in the spatial projection point sequence.

[0089] Step S124: Based on the segment length, inter-segment transitions, and local jump variables, abnormal segments are eliminated to obtain the target travel path of the tracking target within the roadway spatial constraint zone, and the spatial projection points in the target travel path are converted into roadway arc length positions.

[0090] The target roadway progress state is obtained by integrating the spatial projection point and the corresponding roadway arc length position at a unified moment. The roadway arc length position is the cumulative arc length between the spatial projection point and the corresponding roadway centerline reference point in the target passage space constraint model.

[0091] In a specific embodiment, the present invention provides a mobile target localization and tracking method based on unmanned aerial vehicles (UAVs). Step 2, performing continuous spatial projection and sequence construction of the target within a constrained passage zone, includes the following sub-steps:

[0092] Sub-step 2.1: Construct a set of candidate projection points for the target constraint band.

[0093] Specifically, this sub-step does not directly force the target point onto the centerline of the tunnel. Instead, it first finds a set of candidate projection points that satisfy geometric proximity for each target's original spatial location within the tunnel constraint zone. This is because in a port container tunnel scenario, the target may be located in the center of the tunnel, near the edge of the container, or even near the boundary of the constraint zone during short-term occlusion recovery. If only a single centerline projection point is selected from the beginning, it is easy to misjudge the actual edge-hugging motion as center-aligned motion, resulting in unreasonable trajectory compression during subsequent tunnel process reconstruction. Therefore, constructing a set of candidate projection points first and then selecting the optimal point from them is more in line with the engineering implementation logic.

[0094] At each moment, the distance between the original spatial location of the target and the discrete sampling points along the roadway centerline is calculated one by one. Combined with the roadway spatial constraint radius, all candidate points satisfying the in-band constraints are selected. To avoid high-frequency jitter, the spatial proximity, lateral deviation, and height consistency of the candidate points are simultaneously included in the evaluation. The comprehensive distance expression for each centerline sampling point is:

[0095]

[0096] In the formula, Indicates the original spatial location of the target point to the nth The combined distance of all centerline sampling points, in meters; , , Indicates time The horizontal, vertical, and height coordinates of the original spatial location of the target, all in meters; , , Indicates the first The horizontal, vertical, and height coordinates of each centerline sampling point are given in meters. This represents the instantaneous lateral distance from the original spatial location of the target to the centerline of the corresponding tunnel, expressed in meters. Indicates the first The roadway constraint radius at each sampling location is in meters. This represents the height correction factor, which is a dimensionless quantity and ranges from 0.2 to 0.8. This represents the boundary fit correction coefficient, which is a dimensionless quantity with a value range of 0.1-0.5.

[0097] Subsequently, based on the above comprehensive distance, the top 3-8 centerline sampling points with the smallest distance at each time point are selected as candidate projection base points. Then, candidate projection points within the band are generated near these sampling points along the local normal direction, forming a set of candidate projection points for the target constraint band.

[0098] Sub-step 2.2 generates the optimal in-band projection point sequence for the target.

[0099] Specifically, for each time step, the set of candidate projection points is filtered for optimal points. However, the filtering cannot only consider the point closest to the original point, because targets in the port tunnel may experience short-term jumps during local occlusion recovery. If only the principle of instantaneous proximity is applied, projection points will be incorrectly attached to adjacent candidate paths. Therefore, this sub-step incorporates both the point with the minimum current geometric error and the point with continuity with the previous projection position into the objective function, forming the optimal in-band projection point for a single time step.

[0100] In this embodiment, the expression for the optimal projection evaluation value is:

[0101]

[0102] In the formula, Indicates the first The evaluation value of each candidate projection point, in square meters; , , Indicates the first The horizontal, vertical, and height coordinates of each candidate projection point, all in meters; , , This indicates the horizontal, vertical, and height coordinates of the projection point that were determined at the previous moment, all in meters. Indicates the first The lateral deviation of each candidate projection point relative to the central skeleton of the constraint band, in meters; This represents the observation fit weight, which is a dimensionless quantity with a value range of 0.4-0.8; This represents the continuous weight of the time series. It is a dimensionless quantity with a value range of 0.2-0.5. This represents the weight for suppressing lateral sway, which is a dimensionless quantity with a value range of 0.1-0.4.

[0103] Then, the candidate projection point with the smallest evaluation value is selected as the optimal in-band projection point at the current time. This process is repeated for all time points to obtain the target optimal in-band projection point sequence.

[0104] Sub-step 2.3: Target passage sequence function and continuity quality index of projection point sequence.

[0105] Specifically, after obtaining the optimal sequence of projection points arranged in time sequence, this sequence is organized into a target passage sequence. This passage sequence is not simply a collection of points, but rather constructed as an ordered function that expresses the continuous movement of the target along the roadway. That is, the projection points are first sorted by time index, then the segment lengths between adjacent projection points, the transitions between segments, and local jump variables are calculated, and abnormal segments that do not conform to the established roadway passage rules are eliminated. For normal segments, their sequential connection relationships are preserved, forming the target's passage path within the roadway constraint zone.

[0106] In this embodiment, the expression for the continuity quality index between adjacent projection points is:

[0107]

[0108] In the formula, Indicates time The continuity quality index between two adjacent projection segments is a dimensionless quantity. This represents the spatial distance between the projection point at the current moment and the projection point at the previous moment, in meters; This indicates the previous projection distance, in meters. This represents the time interval between the two current projection points, in seconds. This represents the distance normalization constant, in meters, with a value range of 0.2-1.0. This represents the fluctuation penalty coefficient, with units of seconds per meter and a value range of 0.2-1.5. This represents the time smoothing constant, in seconds, with a value range of 0.1-1.0.

[0109] When the aforementioned continuity quality index falls below a preset threshold, the segment is marked as a suspected abnormal segment, and local interpolation repair or breakpoint identification is performed at the end of the current sub-step. After this processing, the target pass sequence function is obtained, which is a sequence of projected trajectories arranged in chronological order and passing the continuity check.

[0110] Sub-step 2.4: Construct the arc length position sequence of the target along the center line of the roadway and the set of the target roadway process states.

[0111] Specifically, based on the passage sequence, the target's position in space is further transformed into where the target has moved along the tunnel. This is because the core challenge in port container tunnel scenarios is not the target's absolute position in three-dimensional space at a single moment, but rather the target's longitudinal movement progress in a highly channelized environment. Therefore, only by transforming the projection points into unified tunnel arc-length coordinates can the true continuity relationship of the target before and after entering the obstructed area be identified.

[0112] First, for each projection point in the target passage sequence, find the corresponding reference point on the roadway centerline. Then, calculate the cumulative arc length from the centerline starting point to that reference point. To improve the accuracy of local bends, the arc length calculation adopts a piecewise accumulation form and incorporates a curvature compensation term, expressed as:

[0113]

[0114] In the formula, Indicates time The target tunnel arc length, in meters; Indicates the time from the start of the tunnel to the current time. The number of centerline segments crossed by the corresponding reference point, which is a positive integer; Indicates the first The length of the centerline sub-segment, in meters. This is the length of the previous centerline sub-segment; The curvature compensation coefficient is a dimensionless quantity with a value range of 0.05-0.25. This formula indicates that the progress of the tracked target in a tortuous tunnel cannot be accumulated solely based on straight segments; the actual path extension caused by boundary bends must also be considered.

[0115] After obtaining the arc length position sequence, the projection point coordinates, arc length position, continuity quality index, and corresponding time are then uniformly organized into a target roadway process state set, which serves as the final output of step 2. This not only preserves the actual projection position of the target within the roadway constraint zone but also provides a standardized process scale of the target in the roadway.

[0116] In this embodiment, step 2 transforms the set of spatial constraint zones obtained in step 1 into a set of target lane process states that can be used for subsequent occlusion segmentation identification through a progressive approach of candidate projection point generation, optimal in-band projection screening, target passage sequence construction, and lane arc length process calculation. The final output is the target optimal in-band projection point sequence, the target passage sequence function, the target arc length position sequence along the lane centerline, and the target lane process state set.

[0117] Step S130: Identify the occlusion interval partitions through the target roadway progress status, and determine the target continuity confidence based on the occlusion interval partitions.

[0118] In some embodiments, the mobile target localization and tracking method based on a drone provided by the present invention includes the following steps in step S130:

[0119] Step S131: Calculate the change in arc length between adjacent arc length positions in the roadway, determine the arc length discontinuity difference between different segments based on the change in arc length, and convert the arc length discontinuity difference into a candidate occlusion trigger value.

[0120] Step S132: Set an initial trigger threshold based on the average distribution of candidate occlusion trigger amounts across multiple sampling intervals, and mark the sampling intervals where the candidate occlusion trigger amounts exceed the initial trigger threshold as candidate occlusion trigger points.

[0121] Step S133: Perform temporal aggregation on candidate occlusion trigger points to merge candidate occlusion trigger points at multiple adjacent times into occlusion interval partitions, and calculate the occlusion intensity corresponding to the occlusion interval partitions.

[0122] In some embodiments, the mobile target localization and tracking method based on a drone provided by the present invention further includes the following steps in step S130:

[0123] Step S134: Obtain the duration of each occlusion interval partition, and filter the occlusion interval partitions according to the duration and occlusion intensity to retain occlusion interval partitions whose duration and occlusion intensity meet the set threshold range.

[0124] The threshold setting should be jointly determined by considering the statistical distribution of candidate occlusion trigger quantities, the rhythm of roadway opening, and the actual occlusion duration pattern, and calibrated using occlusion interval partitioned samples under typical operating conditions. Candidate occlusion trigger quantities should be constructed based on the changes in roadway arc length position at adjacent times, sampling time intervals, and continuity quality indicators; the initial trigger threshold should be adaptively set based on the mean and dispersion of all candidate occlusion trigger quantities, ensuring sufficient sensitivity to periodic, full-column occlusion.

[0125] Step S135: Select effective occlusion intervals from the occlusion interval partitions, and determine the arc length start point and arc length end point corresponding to the effective occlusion intervals, so as to calculate the predicted arc length position of the tracking target at the end point of the effective occlusion interval based on the effective occlusion intervals and the corresponding arc length start point and arc length end point.

[0126] Step S136: Obtain the actual arc length position of the tracking target from the arc length start point to the arc length end point, and calculate the arc length error between the predicted arc length position and the actual arc length position, so as to determine the target continuity confidence based on the arc length error. The arc length error is inversely proportional to the target continuity confidence.

[0127] In a specific embodiment, the present invention provides a mobile target localization and tracking method based on a drone. Step 3, performing periodic occlusion interval segmentation identification and target continuity determination, includes the following sub-steps:

[0128] Sub-step 3.1: Construct the arc length discontinuity difference sequence, the time interval correction sequence, and the candidate occlusion trigger point set.

[0129] Specifically, the target lane process state set obtained in step 2 is first transformed into a discontinuous metric sequence that can be used for segmented identification. This is because the periodic occlusion in port container lanes is not simply the target disappearing once, but rather accompanied by the repeated appearance, disappearance, and reappearance of multiple lane openings. This results in various types of transitions in the arc length sequence: some transitions are genuine occlusion recovery transitions, while others are simply normal process disturbances caused by the target accelerating, decelerating, or adhering to edges locally. Therefore, if a discontinuous metric sequence is not constructed first, and instead a single arc length difference is used for judgment, normal traffic fluctuations will be mixed with genuine occlusion discontinuities, causing distortion in subsequent continuation judgment results.

[0130] First, the arc length change between two adjacent sampling times is calculated to obtain the arc length discontinuity difference. Then, time normalization and continuity quality correction are introduced to form the candidate occlusion trigger value, expressed as:

[0131]

[0132] In the formula, Indicates the first The number of candidate occlusion triggers for each sampling interval, in meters per second; Indicates the first The target arc length position at each moment, in meters. , In order to be with the first The target arc length position at one time is the two arc length positions at adjacent times; Indicates the first The time interval between adjacent sampling intervals, in seconds; Indicates the first The continuous quality index for each interval is a dimensionless quantity. This represents the time smoothing constant, in seconds, with a value range of 0.05-0.5. This represents the continuity penalty coefficient, measured in meters per second, with a value ranging from 0.2 to 1.5. This represents the penalty coefficient for abrupt changes in arc length. It is a dimensionless quantity with a value range of 0.3-1.2.

[0133] After obtaining the candidate occlusion trigger values, an initial trigger threshold is constructed based on the statistical distribution of all sampling intervals. It is recommended that this initial trigger threshold be set using a coupling method of mean and dispersion to ensure sufficient sensitivity to repeated occlusion of roadway openings. Subsequently, all sampling intervals with candidate occlusion trigger values ​​greater than this threshold are marked as candidate occlusion trigger points, forming a set of candidate occlusion trigger points.

[0134] Sub-step 3.2: Construct the set of occlusion start boundaries, the set of occlusion end boundaries, and the set of candidate occlusion intervals.

[0135] Specifically, due to the presence of repeated openings and obstructions in port container tunnels, candidate obstruction trigger points are usually not isolated but appear consecutively in multiple adjacent sampling intervals. Therefore, this sub-step performs temporal aggregation of candidate obstruction trigger points, merging several adjacent trigger points into a single candidate obstruction interval. During aggregation, not only temporal adjacency is considered, but also the consistency of arc length growth direction; if the arc length change direction corresponding to the same set of trigger points is consistent and the occurrence times are close, they are considered components of the same obstruction process.

[0136] In this embodiment, to avoid mistaking extremely short-duration local jumps for complete occlusion regions, it is necessary to quantify the interval intensity of candidate intervals. The expression for interval intensity is:

[0137]

[0138] In the formula, Indicates the first The interval intensity of each candidate occlusion interval, in meters per second; Indicates the first The starting sampling index of each candidate occlusion interval; Indicates the first The sampling index ends for each candidate occlusion interval; This indicates the result of sub-step 3.1. One candidate occlusion trigger quantity, in meters per second; This indicates the time interval for the corresponding sampling interval, in seconds; Indicates the first The total duration of each candidate occlusion interval, in seconds; and These represent the target arc length positions at the time following the end of the candidate interval and at the starting time, respectively, in meters; This represents the interval smoothing constant, in seconds, with a value range of 0.1-1.0; This represents the net process enhancement coefficient of the interval, which is a dimensionless quantity and ranges from 0.2 to 0.8.

[0139] After obtaining the interval intensity, the candidate intervals are filtered according to their intensity and duration, and the intervals that meet the engineering threshold are retained as the candidate occlusion interval set. Then, the first and last moments of each interval are recorded as the occlusion start boundary and occlusion end boundary, respectively, forming the occlusion start boundary set and the occlusion end boundary set.

[0140] Sub-step 3.3 generates a set of effective occlusion intervals and a set of arc length boundary pairs corresponding to the occlusion intervals.

[0141] Specifically, further determination is needed to identify which candidate occlusion intervals are truly valid occlusion intervals exhibiting periodic, continuous occlusion characteristics. This is because some local anomalies in port tunnels, such as a target briefly approaching a container corner fitting, a slight lateral sway in the drone's view, or a spreader passing through a local area, can also create seemingly similar discontinuous intervals. If the candidate intervals are not screened for validity, subsequent continuation determinations will be frequently triggered, increasing the computational burden and making it easier to misclassify normal segments as occluded segments.

[0142] First, the core of effectiveness screening is to examine whether the candidate interval matches the roadway opening rhythm and whether its preceding and following arc length distribution conforms to the structure of continuity before occlusion and reappearance after occlusion. To this end, an occlusion rhythm matching degree needs to be constructed, expressed as:

[0143]

[0144] In the formula, Indicates the first The occlusion rhythm matching degree of each candidate occlusion interval is a dimensionless quantity. This represents the arc length distance between the starting point of the current candidate occlusion interval and the starting point of the previous effective occlusion interval, in meters; This represents the typical opening spacing estimated from the tunnel topology in step 1, in meters; Indicates the first The duration of each candidate occlusion interval, in seconds; , They represent the first The arc lengths corresponding to the start and end points of each candidate occlusion interval are given in meters. , These represent the arc length positions corresponding to the start and end points of the previous candidate occlusion interval, respectively, in meters; This represents the rhythm tolerance constant, expressed in meters, with a value ranging from 0.5 to 3.0. This represents the time reference constant, which is a dimensionless quantity and ranges from 0.2 to 0.6. This represents the duration smoothing constant, in seconds, with a value range of 0.2-1.5. This represents the stability constant of the arc length ratio, in meters, and ranges from 0.1 to 1.0.

[0145] When the matching degree of the occlusion rhythm is higher than the validity threshold, the candidate interval is included in the set of valid occlusion intervals; otherwise, it is removed. The final output is that each valid occlusion interval retains its arc length start point and arc length end point, forming a set of arc length boundary pairs corresponding to the occlusion interval.

[0146] Sub-step 3.4: Construct the target continuity credibility set.

[0147] Specifically, after the effective occlusion interval is identified, a continuity consistency judgment is performed on targets that reappear after the end of each occlusion interval. Continuity consistency does not simply mean looking at how close the newly appearing position is to the predicted position, but rather comprehensively considering the trend before occlusion, the duration of occlusion, and whether the initial advance after reappearance connects with the trend before occlusion. Because multiple adjacent targets often appear sequentially at different openings in port container tunnels, if only the closest distance is considered, it is very easy to mistakenly identify adjacent targets as the original target.

[0148] First, based on the arc length positions of several effective sampling points before occlusion, the average advance rate before occlusion is calculated, and the predicted arc length position is obtained by extrapolating this advance rate and the duration of the occlusion interval. Then, the actual arc length position of the newly appearing target after occlusion is compared with the predicted arc length position to determine the error, and the advance consistency within the first short window after occlusion is further introduced to form the comprehensive continuity reliability, expressed as:

[0149]

[0150] In the formula, Indicates the first The confidence level of the newly emerging target corresponding to each effective occlusion interval is a dimensionless quantity with a value range of 0-1. Indicates the first The position of the first arc length of a newly appearing target after the end of an effective blocking interval, in meters; This indicates the predicted arc length position obtained by extrapolation from the state before occlusion, in meters; This represents the average arc length advance rate of a newly appearing target within the short window after obstruction, expressed in meters per second. This represents the average arc length advance rate within the short window before obstruction, expressed in meters per second. It represents the continuity quality index of a newly emerging target in the first segment after occlusion, and is a dimensionless quantity; This represents the arc length error attenuation constant, in meters, with a value range of 0.5-5.0. This represents the rate error decay constant, expressed in meters per second, with a value ranging from 0.2 to 2.0. This represents the continuous smoothness constant, which is a dimensionless quantity and ranges from 0.05 to 0.3.

[0151] For each valid occlusion interval, the above judgment is performed to obtain the target continuation confidence set. Each element in this set corresponds to the target continuation confidence after the end of an occlusion interval, which is directly used in step 4 for reconstructing the flight path and outputting stable tracking.

[0152] Step S140: Based on the target continuity credibility, the drone escort path is reconstructed, and the target tracking and positioning result is generated by executing the reconstructed drone escort path.

[0153] In some embodiments, the UAV-based mobile target localization and tracking method provided by the present invention includes the following steps in step S140:

[0154] Step S141: Determine the optimal target continuation chain based on the target continuation credibility, and construct a drone escort reference path based on the optimal target continuation chain.

[0155] Step S142: The drone escort path is converged and updated using the drone escort reference path to obtain the reconstructed drone escort path. The optimal target continuation chain and the drone escort path at each moment during the escort process are associated, accumulated, and stored based on the reconstructed drone escort path to obtain the target tracking and positioning result.

[0156] In a specific embodiment, the present invention provides a mobile target localization and tracking method based on unmanned aerial vehicles (UAVs). Step 4, which involves reconstructing the UAV escort path and outputting stable tracking based on continuity reliability, includes the following sub-steps:

[0157] Sub-step 4.1: Determine the optimal target continuation chain number, the current position of the optimal target continuation chain, and the short window stability index of the optimal target continuation chain.

[0158] Specifically, firstly, among the multiple candidate chains given in step 3, the single target continuation chain that should be followed at the current moment is selected. This is because, in the port container tunnel scenario, even if a target continuation credibility set has been formed in step 3, there may still be situations where two or more candidate continuation chains have similar credibility within a local time period. If the maximum credibility at a single moment is used as the sole criterion, the accompanying object is prone to switching back and forth between adjacent moments, causing the drone's path to sway left and right or turn back and forth. Therefore, it is necessary to further superimpose short-window stability and positional continuity on top of the credibility to form the optimal target continuation chain selection value.

[0159] In this embodiment, the expression for the optimal target continuation chain selection value is:

[0160]

[0161] In the formula, Indicates the first The comprehensive selection value of the candidate continuation chains is a dimensionless quantity; This indicates the output of step 3. The continuity confidence of each candidate continuation chain is a dimensionless quantity. Indicates the first The number of consecutive valid samples of a candidate continuation chain within the current short window, which is a positive integer; Indicates the first The spatial distance between the current position of the candidate continuation chain and the current position of the drone, in meters; This represents the average spatial distance between all candidate continuation chains and the drone, in meters. This represents the sampling stability smoothing constant, with a value range of 1-5; This represents the distance penalty coefficient, which is a dimensionless quantity and ranges from 0.3 to 1.5. This represents the distance normalization constant, in meters, with a value range of 0.5-3.0.

[0162] Subsequently, the candidate continuation chain with the largest comprehensive selection value is selected as the optimal target continuation chain, and its horizontal, vertical, and altitude coordinates at the current moment are extracted as the current position of the optimal target continuation chain. Simultaneously, a short-window stability index is constructed based on the continuous effective proportion of the chain's most recent sampling points, providing a basis for smoothing the subsequent escort reference path.

[0163] Sub-step 4.2: Determine the drone's flight reference position, flight reference altitude, and flight reference orientation offset.

[0164] Specifically, after determining the optimal target continuation chain, this sub-step does not directly push the UAV's current position to the target's current position. Instead, it first constructs a suitable accompanying reference position for the port container tunnel scenario. This is because in narrow tunnels, if the UAV gets too close, it is prone to over-approaching due to partial openings at the tunnel top, similar container heights on the side walls, and overlapping operations in adjacent tunnels; if it gets too far away, the path correction will be amplified after the obstruction is removed. Therefore, based on the target's current position, the UAV's current position, and the short-window stability, the accompanying reference position, reference altitude, and reference orientation offset need to be dynamically generated. The higher and more stable the target continuation reliability, the closer the control should be to the target; otherwise, the control should maintain a suitable distance to reserve maneuverability for subsequent reconstruction.

[0165] After obtaining the reference escort distance, the current direction of travel is calculated based on the spatial difference between the two nearest valid positions on the optimal target continuation chain, and an escort reference position is generated in the reverse direction. Simultaneously, an escort reference altitude is given based on the clearance above the tunnel and the target height. If the clearance above the tunnel is small, the reference altitude maintains a small safety margin above the target; if the clearance is large, the reference altitude can be appropriately increased to expand the field of view coverage. Furthermore, to limit the lateral sway of the UAV in the tunnel, a reference orientation offset is constructed. This offset ensures that the UAV's orientation is primarily oriented towards the forward direction of the tunnel centerline, with the target direction as a secondary factor, thereby preventing significant lateral deviation of the UAV's nose when traveling close to the target for a short period.

[0166] Sub-step 4.3 generates the drone's updated flight position, updated flight speed, and updated flight orientation.

[0167] Specifically, this sub-step performs a convergence update on the UAV's current position based on the accompanying reference information generated in sub-step 4.2. Here, a single fixed convergence coefficient cannot be used to directly approximate the reference position because targets in port container tunnels undergo periodic obstruction and recovery. If convergence is performed with the same intensity each time, overshoot is likely to occur after the obstruction ends, manifesting as the UAV suddenly darting forward or suddenly side-swerving. Therefore, the convergence coefficient needs to be jointly adjusted with the current target continuity reliability, short-window stability, and the remaining distance from the UAV to the reference position. That is, it approximates the UAV proportionally according to the remaining deviation from the current position to the reference position, thereby gradually rather than abruptly completing the accompanying path correction, thus forming a dynamic convergence update.

[0168] After the position update is completed, the drone's updated flight speed is calculated based on the position difference before and after the update and the sampling time interval. At the same time, the updated flight orientation is obtained based on the reference orientation offset and the azimuth difference between the current position and the reference position, thus forming a new drone flight state.

[0169] Sub-step 4.4 outputs the target's unique continuous positioning trajectory, the UAV's stable flight path, and the stable tracking results.

[0170] Specifically, this sub-step continuously accumulates and outputs the updated drone escort status. It associates and stores the optimal target continuation chain position at each moment with the updated drone escort position in chronological order, constructing a unique continuous target positioning trajectory and a stable drone escort path. To determine whether the current tracking has reached a stable state, the path volatility within a time window needs to be calculated. A lower path volatility indicates that the drone has not exhibited significant backtracking or jittering during continuous escort. When the path volatility is below a preset threshold and the target continuation reliability remains continuously above a preset lower limit, the output for the current period is determined to be a stable tracking result; otherwise, the status marker indicating that the escort has been updated but is still in the adjustment phase is retained. The final output includes the unique continuous target positioning trajectory, the stable drone escort path, and the stable tracking result status.

[0171] In this embodiment, step 4 follows a progressive approach of selecting the optimal target continuation chain, generating the accompanying reference path, dynamically converging and updating, and outputting stable tracking data. This completes the reconstruction of the UAV accompanying path and the output of stable tracking data based on the continuity credibility. As a result, in the scenario of periodic alignment and occlusion in the port container aisle, the UAV can continuously correct its accompanying path around the unique true continuation target, and finally output the unique continuous positioning trajectory and stable tracking results of the target.

[0172] The following describes a mobile target positioning and tracking device based on an unmanned aerial vehicle (UAV) provided by the present invention. The UAV-based mobile target positioning and tracking device described below and the UAV-based mobile target positioning and tracking method described above can be referred to and correspond to each other.

[0173] like Figure 2 As shown, in one embodiment, a mobile target positioning and tracking device based on an unmanned aerial vehicle (UAV) includes a constraint model construction module, a process state generation module, a target continuity recognition module, and a tracking and positioning output module.

[0174] The constraint model building module is used to acquire container tunnel structure data and perform spatial topology modeling based on the container tunnel structure data to build a target passage space constraint model.

[0175] The process state generation module is used to extract the spatial projection points of the tracking target within the roadway spatial constraint zone from the target access space constraint model, and generate the target roadway process state based on the spatial projection points.

[0176] The target continuity identification module is used to identify the occlusion interval partitions through the target roadway progress status, and to determine the target continuity confidence based on the occlusion interval partitions.

[0177] The tracking and positioning output module is used to reconstruct the drone's flight path based on the target's continuity reliability, and generate the target tracking and positioning result by executing the reconstructed drone flight path.

[0178] The applicant of this invention has provided a detailed description of the embodiments of the invention in conjunction with the accompanying drawings. However, those skilled in the art should understand that the above embodiments are merely preferred embodiments of the invention. The detailed description is only intended to help readers better understand the spirit of the invention and is not intended to limit the scope of protection of the invention. On the contrary, any improvements or modifications made based on the inventive spirit of the invention should fall within the scope of protection of the invention.

Claims

1. A method for locating and tracking a moving target based on an unmanned aerial vehicle (UAV), characterized in that, The method includes: Obtain container tunnel structure data and perform spatial topology modeling based on the container tunnel structure data to construct a target passage space constraint model; Extract the spatial projection points of the tracking target within the lane spatial constraint zone from the target passage space constraint model, and generate the target lane process state based on the spatial projection points; The occlusion interval partitions are identified by the target roadway process status, and the target continuity confidence is determined based on the occlusion interval partitions; Based on the target's continuity credibility, the drone's flight path is reconstructed, and the target tracking and positioning result is generated by executing the reconstructed drone flight path.

2. The mobile target localization and tracking method based on unmanned aerial vehicles according to claim 1, characterized in that, The process of acquiring container tunnel structure data and performing spatial topology modeling based on the container tunnel structure data to construct a target passage space constraint model includes: The drone conducts a cruise scan of the container tunnel area to obtain cruise scan data, and uses a laser rangefinder to determine the edge points of the containers on both sides of the tunnel, so as to generate a tunnel boundary point sequence based on the cruise scan data and the container edge points. The roadway boundary point sequence is subjected to sliding window smoothing fitting and roadway centerline discrete sampling to obtain the roadway boundary curve and roadway centerline discrete sequence.

3. The mobile target localization and tracking method based on unmanned aerial vehicles according to claim 2, characterized in that, The step of acquiring container tunnel structure data and performing spatial topology modeling based on the container tunnel structure data to construct a target passage space constraint model further includes: Multiple sampling sections are selected in the longitudinal direction of the roadway, and the boundary distance between the two sides of the roadway in the sampling section is calculated according to the roadway boundary curve and the discrete sequence of the roadway centerline. The boundary distance is then corrected for fluctuations to obtain the effective passage width of the roadway. The entity width and swing margin of the tracked target and the positioning safety boundary of the UAV are obtained. The effective passage width of the alley is used as a constraint. Based on the entity width and swing margin of the tracked target and the positioning safety boundary of the UAV, the alley space constraint zone is constructed. Based on the discrete sequence of the lane space constraint zone and the lane centerline, the target passage space constraint model is generated.

4. The mobile target localization and tracking method based on unmanned aerial vehicles according to claim 1, characterized in that, The step of extracting the spatial projection points of the tracking target within the roadway spatial constraint zone from the target passage space constraint model, and generating the target roadway progress state based on the spatial projection points, includes: The initial spatial location of the target within the lane spatial constraint zone is obtained, and multiple candidate spatial projection points corresponding to the initial spatial location are selected from the target passage spatial constraint model. The distance between the candidate spatial projection points and the initial spatial location does not exceed a set threshold. Calculate the geometric error between the candidate spatial projection point and the initial spatial position point at the same time, and call the objective function to select the spatial projection point with the smallest function value at the corresponding time based on the geometric error.

5. The mobile target localization and tracking method based on unmanned aerial vehicles according to claim 4, characterized in that, The step of extracting the spatial projection points of the tracking target within the roadway spatial constraint zone from the target passage space constraint model, and generating the target roadway progress state based on the spatial projection points, further includes: The spatial projection points are sorted in chronological order to construct a spatial projection point sequence, and the segment length, inter-segment turning point, and local jump variable between adjacent spatial projection points in the spatial projection point sequence are calculated. Abnormal segments are eliminated based on the segment length, inter-segment transitions, and local jump variables to obtain the target travel path of the tracking target within the spatial constraint zone of the roadway, and the spatial projection points in the target travel path are converted into roadway arc length positions. The target roadway process state is obtained by integrating the spatial projection point and the corresponding roadway arc length position at a unified moment. The roadway arc length position is the cumulative arc length between the spatial projection point and the corresponding roadway centerline reference point in the target passage space constraint model.

6. The mobile target localization and tracking method based on unmanned aerial vehicles according to claim 5, characterized in that, The step of identifying occlusion interval partitions through the target roadway process state and determining the target continuity confidence based on the occlusion interval partitions includes: Calculate the change in arc length between adjacent arc length positions in the roadway, determine the arc length discontinuity difference between different segments based on the change in arc length, and convert the arc length discontinuity difference into a candidate occlusion trigger value. An initial trigger threshold is set based on the average distribution of the candidate occlusion trigger quantity across multiple sampling intervals, and sampling intervals where the candidate occlusion trigger quantity exceeds the initial trigger threshold are marked as candidate occlusion trigger points. The candidate occlusion trigger points are time-series aggregated to merge multiple candidate occlusion trigger points at adjacent times into the occlusion interval partition, and the occlusion intensity corresponding to the occlusion interval partition is calculated.

7. The mobile target localization and tracking method based on unmanned aerial vehicles according to claim 6, characterized in that, The step of identifying occlusion interval partitions through the target roadway process state and determining the target continuity confidence based on the occlusion interval partitions further includes: Obtain the duration corresponding to each occlusion interval partition, and filter the occlusion interval partitions according to the duration and occlusion intensity to retain occlusion interval partitions whose duration and occlusion intensity meet the set threshold range; Valid occlusion intervals are selected from the occlusion interval partitions, and the arc length start point and arc length end point corresponding to the valid occlusion intervals are determined, so as to calculate the predicted arc length position of the tracking target at the end point of the valid occlusion interval based on the valid occlusion intervals and the corresponding arc length start point and arc length end point. The actual arc length position of the tracked target from the arc length start point to the arc length end point is obtained, and the arc length error between the predicted arc length position and the actual arc length position is calculated. The continuity confidence of the target is determined based on the arc length error, and the arc length error is inversely proportional to the continuity confidence of the target.

8. The mobile target localization and tracking method based on unmanned aerial vehicles according to claim 1, characterized in that, The process of reconstructing the UAV escort path based on the target continuity reliability and generating target tracking and positioning results by executing the reconstructed UAV escort path includes: The optimal target continuation chain is determined based on the target continuation credibility, and a drone escort reference path is constructed based on the optimal target continuation chain; The drone escort path is converged and updated using the drone escort reference path to obtain the reconstructed drone escort path. Based on the reconstructed drone escort path, the optimal target continuation chain and the drone escort path at each moment during the escort process are associated, accumulated, and stored to obtain the target tracking and positioning result.

9. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-8.

Images