Low-altitude multi-source heterogeneous detection track threshold cascade fusion method
By using a low-altitude multi-source heterogeneous detection track threshold cascading fusion method, the problems of false association and missed association in scenarios with dense low-altitude targets and equipment switching are solved, and continuous and reliable fused track output for low-altitude targets is achieved.
Patent Information
- Application Number
- CN202611350454.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-02
- Publication Date
- 2026-11-10
AI Technical Summary
Existing multi-source fusion technologies are prone to misassociation, missed association, or track breakage in scenarios with dense low-altitude targets, urban obstruction, short-term data loss, equipment batch number reset, and local coverage switching, making it difficult to maintain stable fusion between overlapping and non-overlapping coverage areas.
The low-altitude multi-source heterogeneous detection track threshold cascade fusion method is adopted. The original observation data of multi-source heterogeneous low-altitude sensing devices are converted into a unified observation model to form local track segments. Multi-dimensional threshold screening, comprehensive scoring and unique allocation are performed at the cross-device level. Combined with backbone drive extension, continuous and reliable fused track segments are output.
To reduce the impact of single-point errors and improve the correlation stability in low-altitude obstruction and dense target scenarios, the output fused track segment set can directly reflect the continuous motion trajectory of low-altitude targets during multi-device coverage switching.
Smart Images

Figure CN123065472A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the fields of low-altitude airspace perception, UAV target detection, multi-source heterogeneous sensor data fusion and real-time trajectory processing, and specifically relates to a low-altitude multi-source heterogeneous detection trajectory threshold cascade fusion method. Background Technology
[0002] Low-altitude surveillance systems typically integrate multiple radar, electro-optical, and laser positioning devices. Different devices may have different output formats, coordinate origins, time bases, observation frequencies, and batch numbering rules. The same low-altitude target often appears as multiple short-term observation segments or localized track segments in different devices.
[0003] Existing multi-source fusion technologies mostly rely on point-level nearest neighbor, statistical distance, or single confidence level fusion. Such solutions are prone to false associations, missed associations, or track breaks in scenarios with dense low-altitude targets, urban obstruction, short-term point loss, equipment batch number reset, and local coverage switching, making it difficult to maintain stable fusion between overlapping and non-overlapping coverage areas. Summary of the Invention
[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a low-altitude multi-source heterogeneous detection track threshold cascade fusion method. This method forms local track segments within each device, and then performs multi-dimensional threshold screening, comprehensive scoring and unique allocation at the cross-device level. It also provides a backbone-driven extension for non-overlapping intervals to output continuous and reliable fused track segments.
[0005] The technical problem addressed by this invention is solved as follows:
[0006] A low-altitude multi-source heterogeneous detection track threshold cascading fusion method includes the following steps:
[0007] Step 1: Convert the raw observation data from multi-source heterogeneous low-altitude sensing devices into a set unified observation model, including observation time, target position, target speed, target heading, source device, device type, device batch number, and observation quality.
[0008] Step 2: For observation data containing the batch number of the same low-altitude sensing device, group them according to the batch number. Within the same group, if the time interval or spatial distance between adjacent observation data exceeds the set segmentation threshold, divide it into new trajectory segments as local trajectory segments. For anonymous observation data without the batch number of the device, calculate the anonymous link cost and connect the anonymous observation data to the local trajectory segment that meets the linking conditions and has the lowest anonymous link cost.
[0009] Step 3: For local trajectory segments of multi-source heterogeneous low-altitude sensing devices, if there is a time overlap between two local trajectory segments, candidate trajectory segment pairs are generated; within the time overlap range, the track points of the two local trajectory segments are paired according to the nearest time principle to generate track pairing point pairs with temporal relationship, and a track pairing point pair set is constructed; calculate the time overlap length, time overlap rate, number of track pairing point pairs, mean position distance, position distance quantile, mean speed difference, and mean heading difference of the track pairing point pair set.
[0010] Step 4: Perform multi-dimensional hard threshold filtering on candidate trajectory segment pairs, including time overlap length threshold, time overlap rate threshold, number of track pairing point pairs threshold, mean position distance threshold, position distance quantile threshold, mean speed difference threshold, and mean heading difference threshold; candidate trajectory segment pairs that pass the multi-dimensional hard threshold are used as candidate fusion trajectory segment pairs;
[0011] Calculate the comprehensive score of candidate fusion trajectory segment pairs, including overlap duration score, number of paired points score, overlap rate score, velocity difference score, heading difference score, position distance quantile value, and average position difference score; candidate fusion trajectory segment pairs whose comprehensive scores exceed a set threshold are selected as trajectory segment pairs to be fused.
[0012] Step 5: For two local trajectory segments with overlapping time ranges, perform track fusion using a weighted fusion method; for two local trajectory segments with non-overlapping time ranges, perform track fusion using a backbone extension method; output the set of fused trajectory segments.
[0013] Furthermore, in step 2, the calculation process for the cost of anonymous links is as follows:
[0014]
[0015] in, This is the current anonymous observation data. Indicates a local trajectory segment. Indicates the current anonymous observation data Link to local trajectory segment The cost of anonymous links; For current anonymous observation data With local trajectory segment Time difference at the end, For current anonymous observation data Relative local trajectory segment Spatial residuals of predicted location For current anonymous observation data and local trajectory segments The speed difference at the end, To normalize the heading difference to [0°, 180°], , and These are the current anonymous observation data. With local trajectory segment The final destination heading , Modulo means taking the modulo, and mod means taking the remainder; , , and These are the weights for time difference, spatial residual, speed difference, and heading difference, respectively. , , and These are the time difference threshold, spatial residual threshold, velocity difference threshold, and heading difference threshold, respectively.
[0016] Furthermore, in step 2, the link condition is set to simultaneously satisfy... , , and If only one local trajectory segment satisfies the link condition, the current anonymous observation data Connect to the candidate local trajectory segment; if multiple local trajectory segments meet the linking conditions, the current anonymized observation data is connected to the local trajectory segment with the lowest anonymized linking cost.
[0017] Furthermore, in step 3, for the two local trajectory segments of low-altitude sensing devices from different sources... and Set of track pairing points Represented as:
[0018]
[0019] in, For local trajectory segments The i-th waypoint and local trajectory segment The track pairing point is formed by the j-th track point; and Local trajectory segments The i-th waypoint and local trajectory segment The time of the j-th track point The maximum allowable pairing time difference for track points within a cross-device trajectory segment;
[0020] Time overlap length for:
[0021]
[0022] Where max represents the maximum value and min represents the minimum value; and These are sets of track pairings. Local trajectory segment and The end time, and These are sets of track pairings. Local trajectory segment and The start time;
[0023] Time overlap rate for:
[0024]
[0025] in, For set Number of track pairing points in China and Local trajectory segments and The number of waypoints;
[0026] Mean distance for:
[0027]
[0028] in, and Local trajectory segments The x and y coordinates of the i-th waypoint and Local trajectory segments The x and y coordinates of the j-th track point;
[0029] Location distance quantile for:
[0030]
[0031] in, This indicates taking the 90th percentile value. For local trajectory segments The i-th waypoint and local trajectory segment The distance between the j-th waypoints;
[0032] Mean speed difference for:
[0033]
[0034] in, and Local trajectory segments The i-th waypoint and local trajectory segment The target velocity at the j-th track point;
[0035] Mean heading difference for:
[0036]
[0037] in, and Local trajectory segments The i-th waypoint and local trajectory segment The target heading at the j-th waypoint.
[0038] Furthermore, in step 4, the time overlap length threshold requirement is: , To set a threshold for time overlap length; the time overlap rate threshold requirement is as follows: , To set a time overlap rate threshold; the number of track pairing point pairs is required to be [number]. , To set a threshold for the number of track pairing points; the average position distance threshold requirement is as follows: , To set the average threshold for location distance; the threshold requirement for location distance quantiles is as follows: , To set the threshold for position distance quantiles; the threshold for the mean velocity difference is required to be... , To set the threshold for the average speed difference; the threshold requirement for the average heading difference is: , To set the threshold for the mean heading difference.
[0039] Furthermore, in step 4, the comprehensive score of the candidate fusion trajectory segment pairs... The calculation process is as follows:
[0040]
[0041] Among them, the average position difference score Location distance quantile Speed difference rating Heading difference score Overlap rate score Overlap duration score Number of matching points score .
[0042] Furthermore, in step 5, the specific process of achieving track fusion through weighted fusion for two local trajectory segments within the time overlap range of the trajectory segments to be fused is as follows:
[0043] Configure basic weights for low-altitude sensing devices from various sources, and determine fusion weights based on the basic weights and observation quality weights; weight the target position, target velocity, and target heading of the two local trajectory segments based on the fusion weights to obtain the fused trajectory segments.
[0044] The specific process of track fusion through backbone extension for two local trajectory segments that do not overlap in time is as follows:
[0045] The non-overlapping time intervals of the trajectory segments to be merged are used as the backbone trajectory segments. Based on the target position, target velocity, and target heading at the boundary time of the backbone trajectory segments, the trajectory is extrapolated to predict the target position, target velocity, and target heading at adjacent candidate prediction times as the predicted trajectory point state. According to the principle of time proximity, the trajectory points of the merged trajectory are selected as candidate extension points, and the target time residual, target position residual, target velocity residual, target heading residual, target lateral residual, and target forward residual are calculated relative to the predicted trajectory point state. When all residuals meet the preset extension threshold, the predicted trajectory point is merged into the merged trajectory.
[0046] Furthermore, in step 5, the fusion weights of the low-altitude sensing device k Represented as:
[0047]
[0048] in, Let k be the basic weight of the low-altitude sensing device. The observation quality weights for low-altitude sensing device k are given by [the following]. The mass ratio is determined by the observation mass weight of the low-altitude sensing device k; min indicates taking the minimum value, and max indicates taking the maximum value. The minimum mass ratio set, This represents the normalized observation quality weight for the low-altitude sensing device k.
[0049] Furthermore, in step 5, for the non-overlapping time intervals in the trajectory segment pair to be fused, let the boundary state of the backbone trajectory segment be... in, These represent the x-axis, y-axis, and z-axis positions in a unified coordinate system. and These are the planar velocity components along the x-axis and y-axis, respectively. For velocity scalar, As the backbone course, The timeframe represents the backbone boundary; adjacent candidate prediction times are... The predicted location is , The target time residual is The residual of the target position is The target velocity residual is The target heading residual is Target forward residual Target lateral residual ,in and The x and y coordinates of the candidate extension points. and For the speed and heading of the candidate extension points;
[0050] The backbone extension threshold is defined as , , , , and ,in, The maximum time residual threshold, The maximum position residual threshold, For the maximum speed residual threshold, The maximum heading residual threshold, This is the maximum lateral residual threshold. The threshold is the maximum forward residual; a candidate extension point is only considered to meet the backbone extension condition if it simultaneously meets all the threshold conditions.
[0051] The beneficial effects of this invention are:
[0052] The method described in this invention employs a local trajectory segment-level association approach to reduce the impact of single-point errors on the fusion results; it improves the association stability in low-altitude occlusion and dense target scenarios through cascaded processing of seven-dimensional hard thresholds and comprehensive scores; it eliminates processing order dependencies through unique allocation; and it supplements single-source extension segments outside the overlapping area and reduces false handovers caused by single-frame noise through backbone-driven extension and buffer confirmation mechanisms; the output fused trajectory segment set simultaneously retains the source device, matching score, and extension status, which can directly reflect the continuous motion trajectory of low-altitude targets during multi-device coverage handover. Attached Figure Description
[0053] Figure 1 This is a flowchart illustrating the method described in this invention;
[0054] Figure 2 This is a schematic diagram of the seven-dimensional threshold and unique allocation in the method described in this invention;
[0055] Figure 3 This is a schematic diagram of the extension of the non-overlapping segment of the backbone drive in the method described in this invention;
[0056] Figure 4This is a comparison image of the trajectory fusion method before and after described in the embodiment;
[0057] Figure 5 The graph shows the performance indicators of the method described in the embodiment. Detailed Implementation
[0058] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0059] This embodiment provides a low-altitude multi-source heterogeneous detection track threshold cascading fusion method, the flowchart of which is shown below. Figure 1 As shown, it includes the following steps:
[0060] Step 1: Convert the raw observation data from multi-source heterogeneous low-altitude sensing devices into a set unified observation model;
[0061] Furthermore, the specific process of step 1 is as follows:
[0062] The latitude and longitude coordinates, local coordinates or polar coordinates of the original observation data are uniformly transformed, the local time, message time or reception time are uniformly transformed into a time axis, and the orientation is corrected according to the orientation parameters of the low-altitude sensing equipment.
[0063] The unified observation model includes observation time, target location, target speed, target heading, source equipment, equipment type, equipment batch number, and observation quality.
[0064] Observation quality is generally obtained by normalizing the quality field provided by the source device. When the source device does not directly provide a quality field but provides signal quality characteristics such as signal-to-noise ratio, peak power, or identification confidence, the device adapter calculates it according to a preset mapping relationship. When there is no reliable quality basis, the observation quality field can be empty.
[0065] Step 2: For the same low-altitude sensing device, construct local trajectory segments based on the continuity of batch numbers within the device and the consistency of motion in anonymous observation data;
[0066] Furthermore, the specific process of step 2 is as follows:
[0067] like Figure 2 As shown, for observation data containing the equipment's batch number from the same low-altitude sensing device, the data is grouped according to the batch number. Within the same group, if the time interval or spatial distance between adjacent observation data exceeds a set segmentation threshold, the data is divided into new trajectory segments as local trajectory segments. For anonymous observation data without an equipment batch number, the anonymous linking cost is calculated based on the time interval, predicted position residual, velocity difference, and heading difference. The anonymous observation data is then connected to the local trajectory segment that meets the linking conditions and has the lowest anonymous linking cost.
[0068] The cost of anonymous links is expressed as:
[0069]
[0070] in, This is the current anonymous observation data. Indicates a local trajectory segment. Indicates the current anonymous observation data Link to candidate local candidate trajectory segment The cost of anonymous links; For current anonymous observation data With local trajectory segment Time difference at the end, For current anonymous observation data Relative local trajectory segment Spatial residuals of predicted location For current anonymous observation data and local trajectory segments The speed difference at the end, To normalize the heading difference to [0°, 180°], and thus avoid erroneous large angle differences when the heading crosses 0° / 360°, and These are the current anonymous observation data. With local trajectory segment The final destination heading , Modulo means taking the modulo, and mod means taking the remainder; , , and These are the weights for time difference, spatial residual, speed difference, and heading difference, respectively. , , and These are the time difference threshold, spatial residual threshold, velocity difference threshold, and heading difference threshold, respectively.
[0071] The link condition is set to meet all conditions simultaneously. , , and If only one local trajectory segment satisfies the link condition, the current anonymous observation data Connect to the candidate local trajectory segment; if multiple local trajectory segments meet the linking conditions, the current anonymized observation data is connected to the local trajectory segment with the lowest anonymized linking cost.
[0072] Step 3: For local trajectory segments of multi-source heterogeneous low-altitude sensing devices, if there is a time overlap between two local trajectory points, candidate trajectory segment pairs are generated; within the time overlap range, the track points of the two local trajectory segments are paired according to the nearest time principle to generate track pairing point pairs with temporal relationship, and a track pairing point pair set is constructed; calculate the time overlap length, time overlap rate, number of track pairing point pairs, mean position distance, position distance quantile, mean speed difference, and mean heading difference of the track pairing point pair set.
[0073] Furthermore, in step 3, for the two local trajectory segments of low-altitude sensing devices from different sources... and Set of track pairing points Represented as:
[0074]
[0075] in, For local trajectory segments The i-th waypoint and local trajectory segment The track pairing point is formed by the j-th track point; and Local trajectory segments The i-th waypoint and local trajectory segment The time of the j-th track point This represents the maximum permissible pairing time difference between track points within a cross-device trajectory segment.
[0076] Time overlap length for:
[0077]
[0078] Where max represents the maximum value and min represents the minimum value; and These are sets of track pairings. Local trajectory segments in the set and The end time, and These are sets of track pairings. Local trajectory segments in the set and The start time.
[0079] Time overlap rate for:
[0080]
[0081] in, For set The number of track pairings in the set and Local trajectory segments and The number of waypoints.
[0082] Mean distance for:
[0083]
[0084] in, and Local trajectory segments The x and y coordinates of the i-th waypoint and Local trajectory segments The x and y coordinates of the j-th track point.
[0085] Location distance quantile for:
[0086]
[0087] in, This indicates taking the 90th percentile value. For local trajectory segments The i-th waypoint and local trajectory segment The distance between the j-th waypoints.
[0088] Mean speed difference for:
[0089]
[0090] in, and Local trajectory segments The i-th waypoint and local trajectory segment The target velocity at the j-th track point.
[0091] Mean heading difference for:
[0092]
[0093] in, and Local trajectory segments The i-th waypoint and local trajectory segment The target heading at the j-th waypoint.
[0094] Step 4: Perform multi-dimensional hard threshold filtering on candidate trajectory segment pairs, including time overlap length threshold, time overlap rate threshold, number of track pairing point pairs threshold, mean position distance threshold, position distance quantile threshold, speed difference threshold, and heading difference threshold; candidate trajectory segment pairs that pass the multi-dimensional hard threshold are used as candidate fusion trajectory segment pairs;
[0095] Calculate the comprehensive score of candidate fusion trajectory pairs, including overlap duration score, number of paired points score, overlap rate score, velocity difference score, heading difference score, position distance quantile value, and average position difference score; candidate fusion trajectory pairs whose comprehensive scores exceed a set threshold are selected as trajectory pairs to be fused.
[0096] Furthermore, in step 4, the time overlap length threshold requirement is: , To set a threshold for time overlap length; the time overlap rate threshold requirement is as follows: , To set a time overlap rate threshold; the number of track pairing point pairs is required to be [number]. , To set a threshold for the number of track pairing points; the average position distance threshold requirement is as follows: , To set the average threshold for location distance; the threshold requirement for location distance quantiles is as follows: , To set the position distance threshold; the velocity difference threshold requirement is... , To set the speed difference threshold; the heading difference threshold requirement is: , To set the heading difference threshold.
[0097] If the time overlap length, time overlap rate, number of track pairings, mean position distance, position distance quantile, mean velocity difference, and mean heading difference of the track pairing set all meet the set threshold requirements, calculate the comprehensive score of the candidate fused trajectory segment pairs. :
[0098]
[0099] Among them, the average position difference score Location distance quantile Speed difference rating Heading difference score Overlap rate score Overlap duration score Number of matching points score The above scoring, together with the hard threshold, forms a cascading structure of "first eliminating obvious errors, then ranking and selecting the best."
[0100] Step 5: For two local trajectory segments with overlapping time ranges, perform track fusion using a weighted fusion method; for two local trajectory segments with non-overlapping time ranges, perform track fusion using a backbone extension method; output the set of fused trajectory segments.
[0101] Furthermore, in step 5, the specific process of achieving track fusion through weighted fusion for two local trajectory segments within the time overlap range of the trajectory segments to be fused is as follows:
[0102] Before performing track weighted fusion, basic device weights are configured for each source low-altitude sensing device. These basic device weights are non-negative coefficients characterizing the relative reliability of the corresponding device's output observation data in the fusion calculation; the larger the basic device weight, the greater the impact of the corresponding device's observation data on the fusion result. Basic device weights can be configured according to device type or specific device identifier, and can be determined or adjusted based on the device's nominal measurement accuracy, field calibration results, historical test or playback errors, and engineering experience.
[0103] The fusion weight is determined based on the weight of the source low-altitude sensing device and the weight of the observation quality. The target position, target velocity and target heading of the two local trajectory segments are weighted based on the fusion weight to obtain the fused trajectory. The target heading is fused using a unit direction vector weighting method.
[0104] In this embodiment, the fusion weights of the low-altitude sensing device k are... Represented as:
[0105]
[0106] in, Let k be the basic weight of the low-altitude sensing device. The observation quality weights for low-altitude sensing device k are given by [the following]. The mass ratio is determined by the observation mass weight of the low-altitude sensing device k; min indicates taking the minimum value, and max indicates taking the maximum value. The minimum mass ratio set, is the normalized observation quality weight for low-altitude sensing device k; the larger the basic weight of the device, the greater the impact of the observation data of the corresponding device on the fusion result; when the source device does not provide reliable observation quality, the quality multiplier is taken as 1.
[0107] Target position and target speed according to Weighted; heading uses circular vectors Weighting is used to avoid distortion of the angle mean near 0° / 360°.
[0108] like Figure 3As shown, the specific process of track fusion through backbone extension for two local trajectory segments that do not overlap in time is as follows:
[0109] The non-overlapping time intervals of the trajectory segments to be merged are used as the backbone trajectory segments. Based on the target position, target velocity, and target heading at the boundary time of the backbone trajectory segments, the trajectory is extrapolated to predict the target position, target velocity, and target heading at adjacent candidate prediction times as the predicted trajectory point state. According to the principle of time proximity, the trajectory points of the merged trajectory are selected as candidate extension points, and the target time residual, target position residual, target velocity residual, target heading residual, target lateral residual, and target forward residual are calculated relative to the predicted trajectory point state. When all residuals meet the preset extension threshold, the predicted trajectory point is merged into the merged trajectory.
[0110] In this embodiment, for the non-overlapping time intervals in the trajectory segment pair to be fused, the boundary state of the backbone trajectory segment is assumed to be... in, These represent the x-axis, y-axis, and z-axis positions in a unified coordinate system. and These are the planar velocity components along the x-axis and y-axis, respectively. For velocity scalar, As the backbone course, The timeframe represents the backbone boundary; adjacent candidate prediction times are... The predicted location is , The target time residual is The residual of the target position is The target velocity residual is The target heading residual is Target forward residual Target lateral residual ,in and The x and y coordinates of the candidate extension points. and The speed and heading of the candidate extension points.
[0111] The strong threshold for keel extension can be defined as follows: , , , , and ,in, The maximum time residual threshold, The maximum position residual threshold, For the maximum speed residual threshold, The maximum heading residual threshold, This is the maximum lateral residual threshold. This is the maximum forward residual threshold. A candidate extension point is only considered to meet the backbone extension condition if it simultaneously satisfies all the threshold conditions.
[0112] If the candidate point is located in a noise hotspot region, a tightening factor can also be introduced. Multiply the corresponding threshold by This reduces the impact of low-altitude fixed false alarm areas on backbone extension. Compared to using only Euclidean range continuation, this method can distinguish between normal forward movement and lateral jumps along the heading, making it suitable for non-overlapping extensions at the boundaries of low-altitude radar, electro-optical, and protocol source coverage.
[0113] In this embodiment, a buffer confirmation mechanism is configured for backbone extension. Multiple track extrapolations are performed, and the predicted track point is only incorporated into the fused trajectory when the residuals corresponding to multiple consecutive candidate extension points all meet the preset extension threshold. If the number of consecutive satisfactions is insufficient or the preset extension threshold is not met midway, the predicted track point is not incorporated into the fused trajectory.
[0114] In this embodiment, the trajectory fusion process is implemented through a unique allocation method, so that any local trajectory segment can participate in at most one fusion result in the same round of cross-device fusion.
[0115] A unique allocation can be represented as The constraints are ,in, Indicates candidate segment pair Overall score Represents a local trajectory segment and Whether it was selected as the final pairing, set the value to 1 if yes, otherwise set it to 0. In the engineering implementation, it can be set to... Greedy unique allocation is performed in descending order, ensuring that any local trajectory segment participates in matching at most once in the same round of cross-device association, thus avoiding a high-quality trajectory segment being repeatedly occupied by multiple short fragments.
[0116] The fused track segment set generated by this invention can be directly output as the continuous sensing result of low-altitude targets. The set describes the continuous motion process of the same low-altitude target in overlapping and non-overlapping coverage areas using unified fused segment identifiers, time-series positions, velocities, headings, source device sets, matching scores, extension states, and quality markers, thus forming a complete closed loop from the original observations to the fused track results.
[0117] In this embodiment, the cross-device trajectory segment pairing parameter can be configured as: minimum time overlap length threshold. Minimum time overlap rate Minimum number of paired points Maximum time difference between pairing points Location distance to mean threshold Threshold of positional distance from quantile value Average speed difference threshold Mean heading difference threshold .
[0118] The backbone extension threshold can be configured as: the maximum time residual threshold. Maximum position residual threshold Maximum speed residual threshold Maximum heading residual threshold Maximum lateral residual threshold Maximum forward residual threshold The number of consecutively confirmed candidate extension points is 3. The above values are an example of an engineering configuration and can be adjusted according to the measurement accuracy of the low-altitude sensing equipment, the target speed, the site coverage scale, and the system output cycle.
[0119] In a set of real flight data, the system receives raw detection data from three low-altitude sensing devices: Radar A, Radar B, and Laser Ranging Device C, and uses the real flight data as reference positions. The raw real flight record contains 5038 lines of data, which, after filtering for valid flight segments, yields 4889 valid reference position points; the three devices together form 1487 raw observation points, including 770 from Radar A, 241 from Radar B, and 476 from Laser Ranging Device C.
[0120] Figure 4 The playback results are represented using local coordinates. One local trajectory segment from radar A is denoted as batch number A1, one local trajectory segment from radar B is denoted as batch number B2, and two local trajectory segments from laser ranging device C are denoted as batch numbers C1 and C2, respectively. These batch numbers are only used to distinguish different local trajectory segments and do not correspond to the actual batch number of the equipment.
[0121] like Figure 4 As shown, the original detection data before fusion contains multiple local trajectory segments formed by different devices and batches. After completing trajectory segment pairing, threshold filtering, comprehensive scoring, and unique allocation according to steps 3 and 4, the overlapping time intervals are weighted and fused according to step 5, and the non-overlapping time intervals are extended along a backbone. After processing, the local trajectory segments corresponding to radar batch A1 and radar batch B2 form fused trajectory 1, and the local trajectory segments corresponding to laser ranging device batches C1 and C2 form fused trajectory 2. Fusion trajectory 1 and fused trajectory 2 are described using a unified fused track identifier, and the source device and source batch information are retained.
[0122] This replay yielded 1370 fused waypoints, of which 517 matched reference positions in the actual flight data. The median position error of these 517 fused waypoints was 0.8 m, and the quantile position error was 2.0 m. Of the 4889 valid reference positions, 85.31% achieved fused waypoint coverage within a 50 m range, and 100% within a 150 m range. Corresponding performance indicators are as follows: Figure 5 As shown.
[0123] The above results show that the present invention can pair, merge and extend local trajectory segments formed by different devices and different batches to form a fused trajectory segment with a unified fused trajectory identifier, which is feasible for engineering implementation.
[0124] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any equivalent substitutions or improvements made to the device type, threshold dimension, scoring weight, unique allocation method, or backbone switching conditions within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A low-altitude multi-source heterogeneous detection track threshold cascading fusion method, characterized in that, Includes the following steps: Step 1: Convert the raw observation data from multi-source heterogeneous low-altitude sensing devices into a set unified observation model, including observation time, target position, target speed, target heading, source device, device type, device batch number, and observation quality. Step 2: For observation data containing the batch number of the same low-altitude sensing device, group them according to the batch number. Within the same group, if the time interval or spatial distance between adjacent observation data exceeds the set segmentation threshold, divide it into new trajectory segments as local trajectory segments. For anonymous observation data without the batch number of the device, calculate the anonymous link cost and connect the anonymous observation data to the local trajectory segment that meets the linking conditions and has the lowest anonymous link cost. Step 3: For the local trajectory segments of multi-source heterogeneous low-altitude sensing devices, if there is a time overlap between two local trajectory segments, candidate trajectory segment pairs are generated; within the time overlap range, the track points of the two local trajectory segments are paired according to the nearest time principle to generate track pairing point pairs with temporal relationship, and a set of track pairing point pairs is constructed. Calculate the time overlap length, time overlap rate, number of track pairings, mean position distance, position distance quantile, mean velocity difference, and mean heading difference of the track pairing set; Step 4: Perform multi-dimensional hard threshold filtering on candidate trajectory segments, including time overlap length threshold, time overlap rate threshold, number of track pairing point pairs threshold, average position distance threshold, position distance quantile threshold, average speed difference threshold, and average heading difference threshold. Candidate trajectory segment pairs with multidimensional hard thresholds are used as candidate fusion trajectory segment pairs; Calculate the comprehensive score of candidate fusion trajectory segment pairs, including overlap duration score, number of paired points score, overlap rate score, velocity difference score, heading difference score, position distance quantile value, and average position difference score; Candidate trajectory segment pairs whose comprehensive scores exceed a set threshold are designated as trajectory segment pairs to be fused. Step 5: For two local trajectory segments with overlapping time ranges, perform track fusion using a weighted fusion method; for two local trajectory segments with non-overlapping time ranges, perform track fusion using a backbone extension method; output the set of fused trajectory segments.
2. The low-altitude multi-source heterogeneous detection track threshold cascading fusion method according to claim 1, characterized in that, In step 2, the calculation process for the cost of anonymous links is as follows: in, This is the current anonymous observation data. Indicates a local trajectory segment. Indicates the current anonymous observation data Link to local trajectory segment The cost of anonymous links; For current anonymous observation data With local trajectory segment Time difference at the end, For current anonymous observation data Relative local trajectory segment Spatial residuals of predicted location For current anonymous observation data and local trajectory segments The speed difference at the end, To normalize the heading difference to [0°, 180°], , and These are the current anonymous observation data. With local trajectory segment The final destination heading , Modulo means taking the modulo, and mod means taking the remainder; , , and These are the weights for time difference, spatial residual, speed difference, and heading difference, respectively. , , and These are the time difference threshold, spatial residual threshold, velocity difference threshold, and heading difference threshold, respectively.
3. The low-altitude multi-source heterogeneous detection track threshold cascading fusion method according to claim 2, characterized in that, In step 2, the link condition is set to simultaneously satisfy... , , and ; If only one local trajectory segment satisfies the link condition, the current anonymous observation data Connect to the candidate local trajectory segment; if multiple local trajectory segments meet the linking conditions, the current anonymized observation data is connected to the local trajectory segment with the lowest anonymized linking cost.
4. The low-altitude multi-source heterogeneous detection track threshold cascade fusion method according to claim 1, characterized in that, In step 3, for the two local trajectory segments of low-altitude sensing devices from different sources... and Set of track pairing points Represented as: in, For local trajectory segments The i-th waypoint and local trajectory segment The track pairing point is formed by the j-th track point; and Local trajectory segments The i-th waypoint and local trajectory segment The time of the j-th track point The maximum allowable pairing time difference for track points within a cross-device trajectory segment; Time overlap length for: Where max represents the maximum value and min represents the minimum value; and These are sets of track pairings. Local trajectory segment and The end time, and These are sets of track pairings. Local trajectory segment and The start time; Time overlap rate for: in, For set Number of track pairing points in China and Local trajectory segments and The number of waypoints; Mean distance for: in, and Local trajectory segments The x and y coordinates of the i-th waypoint and Local trajectory segments The x and y coordinates of the j-th track point; Location distance quantile for: in, This indicates taking the 90th percentile value. For local trajectory segments The i-th waypoint and local trajectory segment The distance between the j-th waypoints; Mean speed difference for: in, and Local trajectory segments The i-th waypoint and local trajectory segment The target velocity at the j-th track point; Mean heading difference for: in, and Local trajectory segments The i-th waypoint and local trajectory segment The target heading at the j-th waypoint.
5. The low-altitude multi-source heterogeneous detection track threshold cascade fusion method according to claim 4, characterized in that, In step 4, the time overlap length threshold requirement is: , To set a threshold for time overlap length; the time overlap rate threshold requirement is as follows: , To set a time overlap rate threshold; The threshold requirement for the number of track pairing points is: , To set a threshold for the number of track pairing points; Location distance mean threshold requirement is , To set the average threshold for location distance; the threshold requirement for location distance quantiles is as follows: , To set the location distance quantile threshold; The threshold requirement for the average speed difference is , To set a threshold for the average speed difference; The mean heading difference threshold requirement is: , To set the threshold for the mean heading difference.
6. The low-altitude multi-source heterogeneous detection track threshold cascading fusion method according to claim 1, characterized in that, In step 4, the comprehensive score of the candidate fusion trajectory segment pairs. The calculation process is as follows: Among them, the average position difference score Location distance quantile Speed difference rating Heading difference score Overlap rate score Overlap duration score Number of matching points score .
7. The low-altitude multi-source heterogeneous detection track threshold cascade fusion method according to claim 1, characterized in that, In step 5, the specific process of achieving track fusion through weighted fusion for two local trajectory segments within the time overlap range of the trajectory segments to be fused is as follows: Configure basic weights for low-altitude sensing devices from various sources, and determine fusion weights based on the basic weights and observation quality weights; weight the target position, target velocity, and target heading of the two local trajectory segments based on the fusion weights to obtain the fused trajectory segments. The specific process of track fusion through backbone extension for two local trajectory segments that do not overlap in time is as follows: The trajectory segments in the non-overlapping time intervals of the trajectory segment pairs to be fused are used as the backbone trajectory segments; the trajectory is extrapolated based on the target position, target velocity and target heading at the boundary time of the backbone trajectory segments, and the target position, target velocity and target heading at the adjacent candidate prediction time are used as the predicted trajectory point state; Based on the principle of time proximity, the track points of the fused trajectories are selected as candidate extension points. The target time residual, target position residual, target velocity residual, target heading residual, target lateral residual, and target forward residual of the candidate extension points relative to the state of the predicted track points are calculated. When all residuals meet the preset extension threshold, the predicted track points are merged into the fused trajectories.
8. The low-altitude multi-source heterogeneous detection track threshold cascading fusion method according to claim 7, characterized in that, In step 5, the fusion weights of the low-altitude sensing device k Represented as: in, Let k be the basic weight of the low-altitude sensing device. The observation quality weights for low-altitude sensing device k are given by [the following]. The mass ratio is determined by the observation mass weight of the low-altitude sensing device k; min indicates taking the minimum value, and max indicates taking the maximum value. The minimum mass ratio set, This represents the normalized observation quality weight for the low-altitude sensing device k.
9. The low-altitude multi-source heterogeneous detection track threshold cascading fusion method according to claim 7, characterized in that, In step 5, for the non-overlapping time intervals in the trajectory segment pair to be fused, let the boundary state of the backbone trajectory segment be... in, These represent the x-axis, y-axis, and z-axis positions in a unified coordinate system. and These are the planar velocity components along the x-axis and y-axis, respectively. For velocity scalar, As the backbone course, The timeframe represents the backbone boundary; adjacent candidate prediction times are... The predicted location is , The target time residual is The residual of the target position is The target velocity residual is The target heading residual is Target forward residual Target lateral residual ,in and The x and y coordinates of the candidate extension points. and For the speed and heading of the candidate extension points; The backbone extension threshold is defined as , , , , and ,in, The maximum time residual threshold, The maximum position residual threshold, For the maximum speed residual threshold, The maximum heading residual threshold, This is the maximum lateral residual threshold. The threshold is the maximum forward residual; a candidate extension point is only considered to meet the backbone extension condition if it simultaneously meets all the threshold conditions.
