Trajectory matching method based on coupling of positioning uncertainty

Abstract

The present application relates to the field of mobile positioning data processing. Trajectories containing position, time and point-level positioning uncertainty are obtained, and spatial coding levels and uncertainty area coverage units are selected according to the uncertainty of each point to generate short-term space-time tokens grouped by source points; during coarse screening, the upper limit of the contribution of each source group is independent of the number of tokens within the group, and candidate window references containing no original coordinates are formed according to double-sided sequence continuous hits; only the point set within the window is verified, and the spatial tolerance is determined by reusing the uncertainty, and points with high uncertainty need to be supplemented with motion evidence or marked as insufficient evidence; and the minimum disclosure summary of the continuously overlapping fragments is output. The present application can suppress the coarse screening score bias caused by error expansion, reduce the number of tokens required for fixed fine-level coverage, and limit the range of verification data.

Description

Technical Field

[0001] This invention relates to the fields of mobile positioning data processing, spatiotemporal databases, and privacy-preserving computing technologies, and particularly to a trajectory coincidence matching method, system, electronic device, and computer-readable storage medium based on positioning uncertainty coupling. Background Technology

[0002] Mobile terminals can continuously collect trajectory points including latitude, longitude, and time. Based on the spatial proximity of two trajectories within a similar timeframe, co-movement, shared routes, or overlapping road segments can be identified. This type of processing can be used for ride-sharing analysis, identification of travel segments, collaborative analysis of moving objects, and authorized location services.

[0003] Actual location points typically also carry horizontal positioning accuracy or other positioning quality information. Due to satellite obstruction, urban canyons, equipment differences, sampling strategies, and motion conditions, the error range of different location points can vary significantly. If the same spatial grid and fixed distance threshold are applied to all location points, a fine grid is prone to missing recalls at the boundaries, while a coarse grid introduces a large number of irrelevant candidates; a fixed distance threshold may also misclassify brief intersections, reverse passages, or adjacent parallel roads as parallel roads.

[0004] To reduce the computational cost of comparing all pairwise trajectories, existing schemes often use spatiotemporal grids, multi-level indexes, hashing, or Bloom filters for coarse screening of candidates, followed by precise coordinate or sequence comparisons. Other schemes adaptively select the spatial coding level based on the geometry of the trajectory or trajectory segments, or sort, merge, and precisely match candidate similar segments.

[0005] However, if a location point is extended to multiple adjacent spatial units to tolerate positioning errors, and the number of shared tokens is directly accumulated, then locations with larger errors may generate more tokens, thus obtaining higher scores in the initial screening. This score increase comes from representing quantity rather than evidence of actual common movement, forming a structural bias.

[0006] Furthermore, if coarse screening and fine testing use unrelated spatial parameters, inconsistencies may arise between the coverage area of ​​coarse screening and the distance tolerance of fine testing. For example, if coarse screening recalls candidates based on regions with large errors, while fine testing still uses a fixed small threshold, it will result in missed detections; if coarse screening uses a fixed fine grid, while fine testing uses a large dynamic tolerance, candidates that could have entered fine testing will be lost in the coarse screening stage.

[0007] If the candidate index service returns or retrieves the candidate's complete original trajectory after a token hit, it expands the scope of location data exposure. Simply pruning the start and end points in the final results stage does not reduce the exposure of the original trajectory that has already occurred during the candidate generation and refinement stages.

[0008] Therefore, a trajectory overlap matching scheme is needed to create a clear data dependency between candidate representations and fine verification for the same positioning uncertainty, avoid artificially increasing the coarse screening score by expanding the number of tokens, and limit the fine verification data to the local window corresponding to continuous candidate evidence. Summary of the Invention

[0009] The purpose of this invention is to provide a trajectory coincidence matching method based on positioning uncertainty coupling, so as to solve the technical problems of disconnection between candidate representation and precision judgment under inconsistent positioning error conditions, coarse screening bias caused by the number of extended tokens, and excessive exposure range of the original trajectory in the candidate stage.

[0010] To achieve the above objectives, one embodiment of the present invention includes the following steps.

[0011] S101, acquire the first trajectory point sequence and the second trajectory point sequence. Each trajectory point includes at least its location, acquisition time, and point-level positioning uncertainty. The point-level positioning uncertainty can be derived from the horizontal positioning accuracy field, or it can be obtained from at least one of the following: positioning source, vertical accuracy, sensor status, map matching residual, and historical quality.

[0012] S102 performs validity checks, time sorting, outlier handling, trajectory segmentation, and optional resampling on the trajectory points to form standardized trajectory points. Standardized trajectory points retain original quality evidence, derived uncertainties, and reasons for deletion or downweighting.

[0013] S103, for each standardized trajectory point p_i, an uncertain region U_i is formed with position x_i and point-level positioning uncertainty a_i, and at least one spatial coding level L_i is determined based on a_i. Under each selected level r, a set of spatial cells C_i(r) that at least partially intersects with U_i is determined.

[0014] In one implementation, the finest level whose unit scale satisfies s(r) ≥ γ × a_i is selected as the master level from multiple candidate spatial coding levels, where s(r) represents the unit feature scale of level r and γ is a positive configuration parameter; a predetermined number of adjacent coarser levels can also be selected. The level can also be determined by lookup tables, error levels, traffic modes, or spatial density.

[0015] The uncertain region U_i can be a circle or polygon centered at x_i with radius a_i or a function of a_i, or it can be a confidence ellipse, probability isotropic region, or map constraint region provided by the positioning system. Spatial coding can employ hexagonal hierarchical grids, spherical hierarchical grids, Geohash, Hilbert coding, quadtrees, or other coding methods that can determine the intersection relationships of spatial units.

[0016] When the number of spatial units intersecting with U_i exceeds the resource limit, a coarser level can be selected, the coverage truncation quality mark can be recorded, the coarse screening weight of the point can be reduced, or the point can be marked as not meeting the quality conditions. Silent truncation without recording the quality status should not be regarded as a complete coverage.

[0017] S104, generate spatiotemporal tokens based on spatial units, corresponding time buckets, and policy versions. Spatiotemporal tokens generated from the same standardized trajectory point in one or more spatial coding levels, spatial units, and adjacent time buckets are associated as token groups G_i from the same source point.

[0018] In one implementation, the spacetime token is generated using the following formula: token = Truncate(HMAC(K_epoch, Encode(schemaVersion, epochID, resolutionLevel, spatialCellID, timeBucketID))). Here, K_epoch is a key with an expiration date, Encode is a deterministic canonical encoding, and Truncate is the truncation length according to the configuration. Account identifiers are not included in the encoding input of the comparable token.

[0019] Spacetime tokens can also be generated through other keyed one-way functions, unintentional pseudo-random functions, tokenization processes within trusted execution environments, intersection inputs of private sets, or representations of spatial features that cannot be directly deduced. Ordinary keyless hashes, because they may be enumerated, are only used as a contrastive implementation that does not involve privacy requirements.

[0020] The source point token group G_i is associated with at least one of the following: source group identifier, temporary track identifier, track point sequence number, policy version, and expiration date. The source group identifier remains stable within the corresponding temporary track and key period, but a long-term account identifier is not used.

[0021] S105, the candidate index role establishes a spatiotemporal token inverted index and performs matching on the source point token groups of the first and second trajectory point sequences. Shared spatiotemporal tokens are first aggregated within the source point token groups, and then aggregated at the trajectory level.

[0022] For any source point token group G_i, its group contribution GroupContribution(G_i) satisfies 0≤GroupContribution(G_i)≤M, where M is a predetermined upper limit that does not change with the number of spatiotemporal tokens contained in G_i. Thus, adding adjacent spatial units or time bucket tokens will not cause a single location point to increase its coarse-screen contribution indefinitely.

[0023] In the first implementation, as long as the source token group has a shared spatiotemporal token, its contribution is min(M, matchedTokenCount). When M=1, a contribution cap is formed for the source group. In the second implementation, the contribution is matchedTokenCount / max(tokenCount(G_i), tokenCount(G_j)). In the third implementation, weights are assigned to each token within the group, and the sum of the weights is not greater than M. The contribution of shared tokens is calculated based on the sum of the weights.

[0024] The trajectory-level coarse screening score can be a weighted average or weighted sum of the contributions of token groups from each source point within a continuous running segment. The weights can be determined based on positioning quality, time difference, spatial hierarchy, motion state, and token policy, but the upper limit of the contribution of a single source point token group does not increase with the number of its tokens.

[0025] S106, a candidate window reference is formed based on the sequential group-level hits in the two trajectories. The candidate window reference includes at least the start and end sequence ranges of the first trajectory point sequence and the second trajectory point sequence, but does not include the original coordinate sequences of the two trajectories.

[0026] Specifically, based on the sequence positions in the first trajectory point sequence, group-level hits are arranged, and continuous running segments in the second trajectory point sequence are selected where the sequence positions remain monotonic and the sequence gaps, time gaps, or travel distance gaps between adjacent hits do not exceed the corresponding thresholds. When a running segment meets the minimum group number, duration, or travel distance conditions, a candidate window reference is generated.

[0027] Candidate window references may also include temporary trajectory identifiers on both sides, number of matching groups, coarse screening score, first and last time buckets, strategy version, quality flag, and expiration time. The candidate index role only outputs candidate window references and limited coarse screening evidence, and does not obtain or output the candidate's complete original trajectory for a single candidate hit.

[0028] S107, the authorized data retrieval role obtains the first and second candidate point sets based on the candidate window reference, and then assigns the limited point set to the restricted verification role. The authorized data retrieval role can be located in a mobile terminal, a permission-isolated service, or a trusted execution environment.

[0029] S108, the point-level positioning uncertainty on which the restricted verification role is used to generate the corresponding source point token group, determines the spatial matching tolerance of candidate point pairs, and performs fine verification in conjunction with the time sequence and continuity of data collection.

[0030] For point p_i in the first candidate point set and point q_j in the second candidate point set, the spatial matching tolerance can be determined by the following formula: R_ij=Clamp(R_min, R_max, β_0+β_1×sqrt(a_i^2+a_j^2)). Where a_i and a_j are point-level positioning uncertainties of the same origin as the token phase, and β_0, β_1, R_min and R_max are scene configuration parameters.

[0031] Spatial compatibility can be taken as max(0, 1-distance(p_i, q_j) / R_ij), or it can be taken as the overlap rate of uncertain regions, the expected distance, the probabilistic distance, or other calibrated monotonic functions. Point-level positioning uncertainty does not need to be assumed to be the standard deviation of a specific normal distribution.

[0032] Before fine-grained verification, the quality status is determined based on point-level location uncertainty and optional quality evidence. Points that meet the first quality condition proceed to dynamic tolerance verification; points that do not meet the first quality condition cannot be considered a positive match solely based on the overlap of two uncertain regions.

[0033] Points that do not meet the first quality condition can be marked as insufficient evidence, excluded, or have their weight reduced, or the direction, speed, road segment, perpendicular position, mode of transportation, or other independent evidence can be required to meet the reinforcing condition. When the quality is restored, the continuous segment can be restarted.

[0034] Time compatibility can be determined based on the time difference between two acquisition points and a time threshold. Direction can be obtained from the displacement vectors of multiple adjacent trajectory points; when displacement is insufficient or velocity is too low, the reliability of the direction is reduced. Hard exclusion can be performed if the direction is reliable but continuously reversed.

[0035] Additional compatibility can be determined using speed difference, mode of transport compatibility, map-matched road segments, road direction, altitude, vertical accuracy, or barometric pressure variations. This evidence is used only when the corresponding input is available, and is weighted and normalized according to the available evidence.

[0036] Two candidate trajectories with different sampling rates can be aligned using time-monotonic dynamic programming, sliding window, hidden Markov model, graph matching, or constrained common subsequence algorithms. Run segments are segmented when the time, sequence, or distance gap between adjacent valid point pairs exceeds a threshold.

[0037] S109: Extract consecutive overlapping segments from the detailed verification results that meet the minimum number of valid points, duration, or travel distance conditions. Brief road intersections, continuous reversals, long-term time misalignments, or consecutive low-compatibility points do not form qualified overlapping segments.

[0038] Before output, boundary privacy processing is performed on consecutive overlapping segments. Boundary privacy processing includes cutting off a predetermined time or distance from both ends of the segment, mapping the boundary to coarser spatial units and time buckets, expanding the cutting range based on the risk of residence, school, workplace, or low-frequency location, and outputting at least one of the following if the processed segment is less than the disclosure threshold: outputting only the result of overlap or not outputting the result.

[0039] The overlap digest can include distance intervals, duration intervals, confidence levels, coarse-grained regions, and expiration times, but does not include the complete sequence of overlap coordinates. Candidate window verification data is deleted after computation is complete or after a timeout. Online tokens expire with the key cycle, and the business layer only saves the digest within the authorized scope.

[0040] This invention also provides a trajectory overlap matching system, including a trajectory preprocessing module, a tokenization module, a candidate index module, a restricted verification module, and a digest module. Each module can be deployed on different physical devices, or they can be logically isolated within the same device through keys, access permissions, processes, and data retention policies.

[0041] The present invention also provides an electronic device, including a processor and a memory, wherein a program in the memory, when executed by the processor, implements the above-described method. The present invention also provides a non-transitory computer-readable storage medium storing the program.

[0042] Compared with the prior art, the present invention has at least the following beneficial effects: First, the same positioning uncertainty runs through spatial hierarchy selection, uncertainty region coverage and precision tolerance, reducing the disconnect between coarse screening and precision parameters; Second, the contribution limit of the source point token group is independent of the number of tokens, suppressing the artificial increase in scores of high error points due to the expansion of more tokens; Third, the candidate index role only outputs bilateral continuous window references, limiting the range of data entering the precision test; Fourth, quality gating avoids the formation of a positive match only due to the overlap of excessively high uncertainty regions. Attached Figure Description

[0043] Figure 1 This is a flowchart illustrating the overall process of trajectory overlap matching according to one embodiment of the present invention.

[0044] Figure 2 This is a schematic diagram of system roles and data boundaries according to one embodiment of the present invention.

[0045] Figure 3 This is a schematic diagram illustrating the generation of uncertainty-driven spatial hierarchy and covering unit according to one embodiment of the present invention.

[0046] Figure 4 This is a schematic diagram illustrating source point token group normalization and continuous candidate window generation according to one embodiment of the present invention.

[0047] Figure 5This is a flowchart illustrating the fine verification and quality gating process of one embodiment of the present invention.

[0048] Figure 6 This is a schematic diagram illustrating the minimum disclosure and data lifecycle of one embodiment of the present invention. Detailed Implementation

[0049] The present invention will be further described below with reference to the accompanying drawings and embodiments. The embodiments are used to explain the technical solutions, not to limit the scope of protection. Parameters not specifically described can be configured based on the location source, mode of transportation, map environment, computing resources, and calibrated experimental results.

[0050] Example 1 employs a mobile terminal, tokenized roles, candidate index services, authorized data retrieval roles, restricted verification services, and business services. The mobile terminal collects raw location points, which include latitude, longitude, timestamp, and horizontal positioning accuracy, and may also include speed, heading, altitude, vertical accuracy, and mode of transportation.

[0051] The mobile terminal deletes illegal coordinates, non-positive positioning accuracy, duplicate points, and points in reverse time sequence; it marks abnormal points based on point age, upper speed limit, and spatial jumps; and it segments the trajectory based on time gaps, pause states, or changes in motion mode. The horizontal positioning accuracy, after being processed with upper and lower limits, is used as the point-level positioning uncertainty a_i.

[0052] This embodiment uses an H3-level grid as an example of spatial coding, with candidate levels 8, 9, 10, and 11. The targetEdge_i = 1.5 × a_i is calculated, and the finest candidate level with an average hexagonal side length not less than targetEdge_i is selected as the primary level. At most one adjacent coarser level is also selected. The grid type, level, and coefficients are all replaceable.

[0053] Using a 24-sided polygon as the center and a_i as the radius, approximate the uncertain circle and find the H3 cells that intersect with this polygon at each selected level. At least one center cell containing x_i must be included. Set an upper limit on the number of cells per level; record a truncation flag when the upper limit is reached.

[0054] Each layer corresponds to a different time bucket granularity. For example, levels 8, 9, 10, and 11 of H3 can correspond to 300-second, 120-second, 60-second, and 30-second time buckets, respectively, and can generate features of adjacent time buckets to handle bucket boundaries. The time lengths mentioned above are examples and not fixed limitations.

[0055] A source point token group is generated for each standardized trajectory point. The token input includes the policy version, key period, spatial level, spatial unit, and time bucket, truncated to 128 bits using HMAC-SHA256. The source group identifier is generated in a separate domain from the temporary trajectory identifier and trajectory point sequence number; the candidate index service does not hold the token key.

[0056] The candidate index service builds an inverted index based on token values. During a query, tokens shared by the same query source group and candidate source group are counted first, and then the contribution of that group is capped at 1. For each query sequence position, the candidate group with the highest contribution and that meets time or order constraints is selected, forming a hit sequence where the sequence positions on both sides remain monotonic.

[0057] This embodiment requires that the continuous running segment contains at least three source groups and allows for a limited number of sequence gaps. The candidate window reference records the query temporary trajectory identifier, candidate temporary trajectory identifier, query start and end sequences, candidate start and end sequences, number of matching groups, coarse screening score, and strategy version. This object does not contain latitude, longitude, or trajectory point arrays.

[0058] Authorized data retrieval roles extract the corresponding point set only based on candidate window references. The restricted verification service first performs quality gating. In this embodiment, 75 meters is used as the experimental threshold for the first quality condition; points with uncertainty higher than this value do not participate in affirmative segments based solely on location, time, and continuity, requiring other independent motion evidence. The production threshold should be calibrated based on actual data.

[0059] For candidate point pairs that meet the quality conditions, with R_min = 10 meters, R_max = 200 meters, β_0 = 8 meters, and β_1 = 1, the spatial tolerance is calculated using R_ij = Clamp(R_min, R_max, β_0 + β_1 × sqrt(a_i^2 + a_j^2)). If the time difference between two points exceeds 45 seconds, they are not considered candidate point pairs; otherwise, the spatial compatibility is calculated based on the distance-to-tolerance ratio.

[0060] The candidate point pairs are kept monotonic in time and sequence, allowing a finite gap of one or two points. An initial segment is formed after three consecutive valid point pairs are reached; segmentation occurs when there is continuous reversal, temporal misalignment, or insufficient quality. In practice, further evidence of direction, velocity, road segmentation, and vertical position can be added.

[0061] After successful verification, the configured distance or time is removed from both ends of the continuous segment, and the distance, duration, and confidence level are output as an interval or discrete level. The verification service deletes the window point set, the indexing service deletes tokens periodically, and the business service does not accept complete coincident coordinate sequences.

[0062] To verify the technical relationship, ten scenarios were constructed using a fixed synthetic trajectory, including co-trajectory, different sampling rates, mid-section drop-off, grid boundary offset, high-noise co-trajectory, brief intersection, reverse, parallel roads, time misalignment, and high-uncertainty parallel trajectory. Each scenario was run thirty times.

[0063] In the synthetic experiment, the fixed H3 level 11 error coverage scheme and the uncertainty coupling capping scheme yielded the same classification results. Compared to the fixed fine-level scheme, the coupling scheme reduced the average number of tokens on the query side from 290.02 to 179.67 and on the candidate side from 353.05 to 217.31, representing reductions of 38.05% and 38.45%, respectively. These figures only illustrate the effect of this embodiment on fixed synthetic data and do not represent the actual deployment accuracy.

[0064] When the uncertainty at the same point increases from 5 meters to 120 meters, the number of extended token groups increases from 6 to 15. If the original count of shared tokens is used directly, the contribution at a single point increases synchronously. After adopting the source group contribution capping, the contribution of each source group does not exceed 1, thus eliminating the problem of the contribution cap increasing with the number of extended tokens from the data structure perspective.

[0065] In the first batch of experiments, two parallel trajectories, each with an uncertainty of 80 meters and an interval of approximately 120 meters, were misclassified. After implementing a high uncertainty insufficient evidence gating, this type of scenario did not form a definite match in the second batch of experiments. This failure and correction demonstrate that quality gating and dynamic tolerance need to be used in conjunction.

[0066] Example 2 employs an intra-group proportional contribution. For source groups G_i and G_j, the contribution is set to matchedTokenCount / max(tokenCount(G_i), tokenCount(G_j)) and limited to between 0 and 1. This method reduces the contribution of groups with only a small number of overlapping coverage units, and shrinks the average proportion of the detection window in the synthesis experiment. However, it may result in a small number of missed detections in high-noise co-channel scenarios. Therefore, either a capping or proportional approach can be selected based on recall priority or window shrinkage priority.

[0067] Example 3 uses probabilistic quality normalization. The tokenization role assigns weights to tokens based on the overlap ratio between the uncertain region and each spatial unit, ensuring that the weights of the same source group sum to 1. Candidate index services accumulate the minimum weight of shared tokens or other bounded combinations, maintaining the single-group contribution ceiling independent of the number of tokens.

[0068] Example 4 deploys the verification on the user terminal. The candidate index service sends candidate window references to authorized terminals. The two terminals exchange or compare limited window features via an encrypted protocol, or one terminal obtains the protected window representation of the other. The terminal outputs a digest, and the server does not obtain the original window coordinates.

[0069] Example 5 deploys verification within a trusted execution environment or a secure multi-party computation protocol. Token indexing, key holding, and verification roles remain logically isolated, and candidate window references continue to limit the scope of protocol input.

[0070] Example 6 uses direction and road segment reinforcement for high uncertainty points. High uncertainty points only participate in affirmative segments if the direction reliability reaches a threshold, the direction difference is less than a threshold, and the map-matched road segments are the same or topologically adjacent. For elevated, ground-level, or underground roads of different levels, vertical position evidence compatibility can also be required.

[0071] Example 7: Adaptive boundary handling based on sensitive location risk. When the beginning and end of consecutive overlapping segments fall near residences, schools, workplaces, or low-frequency locations, the cut-off distance or time is increased; if the remaining segments are less than the minimum disclosure threshold, only the existence of overlap is output or no result is output.

[0072] Those skilled in the art will understand that, without departing from the core data relationships of this invention, spatial encoding, token functions, bounded aggregation within groups, continuous running segment algorithms, precise alignment algorithms, and deployment roles can all be replaced with equivalent alternatives. The scope of protection of this invention is defined by the claims.

Claims

1. A trajectory coincidence matching method based on positioning uncertainty coupling, characterized in that, include: A first trajectory point sequence and a second trajectory point sequence are obtained, wherein each trajectory point includes at least its location, acquisition time, and point-level positioning uncertainty. For each trajectory point, at least one spatial coding level and a set of spatial units intersecting with the uncertain region of the trajectory point under the spatial coding level are determined based on the point-level positioning uncertainty of the trajectory point. Spatiotemporal tokens are generated based on the set of spatial units and the corresponding time buckets, and multiple spatiotemporal tokens generated from the same trajectory point are associated as the same source point token group. The source point token groups of the first trajectory point sequence and the second trajectory point sequence are matched through candidate index roles. When aggregating shared spatiotemporal tokens, the coarse screening contribution upper limit of a single source point token group is independent of the source point token group. The number of spatiotemporal tokens contained in the deck is used to form candidate window references based on the sequential group-level hits in the first trajectory point sequence and the second trajectory point sequence. The candidate window references include the sequence range of the two trajectories but do not include the original coordinate sequences of the two trajectories. The first candidate point set and the second candidate point set within the corresponding sequence range are obtained based on the candidate window references. The spatial matching tolerance of the candidate point pairs is determined by reusing the point-level positioning uncertainty used when generating the corresponding source point token group. The first candidate point set and the second candidate point set are finely verified based on the spatial matching tolerance, the acquisition time sequence and the continuity condition to obtain continuous overlapping segments. And an overlapping summary that does not contain a complete overlapping coordinate sequence is output.

2. The method according to claim 1, characterized in that, The step of determining at least one spatial coding level and a set of spatial cells based on point-level positioning uncertainty includes: converting the point-level positioning uncertainty into an uncertain region based on the location of the trajectory point; selecting the finest spatial coding level from multiple candidate spatial coding levels whose cell scale satisfies the coverage condition of the uncertain region, and selecting at most a predetermined number of adjacent coarser spatial coding levels; determining spatial cells in each selected spatial coding level that at least partially intersect with the uncertain region; and recording a coverage truncation quality marker when the number of spatial cells exceeds the resource limit.

3. The method according to claim 1, characterized in that, The spatiotemporal token is obtained by performing a keyed one-way function on the normalized coding features of a key pair with a validity period. The normalized coding features include at least a policy version, a key period identifier, a spatial coding level, a spatial unit identifier, and a time bucket identifier. The account identifier is not a component field of the normalized coding features. The source point token group is also associated with a source group identifier and a trajectory point sequence number that are valid within the range of the temporary trajectory identifier.

4. The method according to claim 1, characterized in that, The coarse screening contribution of a single source point token group is determined in any of the following ways: when the source point token group has a shared spatiotemporal token, the contribution of the source point token group is capped to a predetermined constant; the number of shared spatiotemporal tokens is divided by the token number normalization factor of the two source point token groups participating in the matching. Alternatively, assign weights to each spatiotemporal token within the source point token group and ensure that the sum of the weights within the group does not exceed a predetermined constant.

5. The method according to claim 1, characterized in that, The process of forming a candidate window reference includes: arranging group-level hits according to the sequence positions in the first trajectory point sequence; selecting continuous running segments in the second trajectory point sequence where the sequence positions remain monotonic and the gaps on both sides do not exceed the corresponding gap thresholds; generating a candidate window reference when the continuous running segment reaches at least one of the group number threshold, duration threshold, or travel distance threshold; the candidate window reference also includes at least one of the following: temporary trajectory identifier, start and end sequence positions on both sides, number of matching groups, coarse screening evidence, strategy version, and expiration time.

6. The method according to claim 1, characterized in that, For point p_i in the first candidate point set and point q_j in the second candidate point set, the spatial matching tolerance R_ij is determined according to the following formula: R_ij=Clamp(R_min, R_max, β_0+β_1×sqrt(a_i^2+a_j^2)), where a_i and a_j are the point-level positioning uncertainties used to generate the source point token groups corresponding to point p_i and point q_j, respectively, R_min and R_max are the upper and lower limits of the tolerance, and β_0 and β_1 are configuration parameters; and the spatial compatibility is determined according to the ratio of the distance between the two points to R_ij.

7. The method according to claim 1, characterized in that, Before fine verification, the quality status is determined based on the point-level positioning uncertainty. When the point-level positioning uncertainty of a candidate point does not meet the first quality condition, not only is a positive match formed based on the overlap of the uncertainty areas of two candidate points, but at least one of the following is performed: marking the candidate point as insufficient evidence, reducing the matching weight of the candidate point, excluding the candidate point, or requiring at least one independent piece of evidence from direction, speed, road segment, vertical position, and mode of transportation to meet the reinforcement condition.

8. The method according to claim 1, characterized in that, The refined verification also includes: determining the direction and direction reliability based on the displacement vectors of multiple adjacent trajectory points; when the direction reliability meets the direction usage conditions, determining the direction compatibility based on the direction difference and excluding continuous reverse motion; determining additional compatibility based on at least one of time difference, speed difference, traffic mode compatibility relationship, map matching road segment or vertical position; and extracting continuous overlapping segments by dynamic programming, sliding window, hidden Markov model, graph matching or constrained common subsequence algorithm that keeps the acquisition time monotonic.

9. The method according to claim 1, characterized in that, Before outputting the overlapping digest, at least one of the following boundary privacy processing is performed on the continuous overlapping segments: cutting off a predetermined time or distance from both ends, mapping both ends to coarser spatial units and time buckets, expanding the cutting range according to the risk of sensitive locations, or only outputting the existence of overlap or not outputting the result when the processed segment does not meet the disclosure threshold; the candidate index role does not hold the token key, the candidate window verification data is deleted after verification is completed or timeout, and the spatiotemporal token expires according to the key period.

10. A trajectory coincidence matching system based on positioning uncertainty coupling, characterized in that, include: The trajectory preprocessing module is used to obtain a sequence of trajectory points that includes location, acquisition time, and point-level positioning uncertainty; The tokenization module is used to determine the spatial coding level and uncertain region coverage unit based on the point-level positioning uncertainty of each trajectory point, and generate source point token groups; The candidate index module is used to aggregate shared spatiotemporal tokens by source point token groups, making the coarse screening contribution limit of a single source point token group independent of the number of spatiotemporal tokens within the group, and outputting candidate window references without the original coordinate sequence based on the sequential group-level hits in the trajectories on both sides; the restricted verification module is used to obtain only the point set corresponding to the candidate window references, reuse the corresponding point-level positioning uncertainty to determine the spatial matching tolerance, and obtain continuous overlapping segments by combining the acquisition time sequence and continuity conditions; and the summary module is used to output overlapping summaries that do not contain complete overlapping coordinate sequences.

11. An electronic device comprising a processor and a memory, wherein the memory stores a computer program, characterized in that, When the computer program is executed by the processor, it causes the electronic device to perform the method of any one of claims 1 to 9.

12. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 9.

Images