Vehicle trajectory reconstruction and correction method and device based on spatiotemporal data fusion

Abstract

The embodiment of the application provides a vehicle trajectory reconstruction and deviation correction method and device based on space-time data fusion, innovatively designs a multi-source data fusion model, realizes unified processing of data through coordinate conversion and time calibration, constructs a trajectory completion system, combines road network features and real-time road conditions, establishes a reliable trajectory reconstruction mechanism, introduces visual editing, and ensures the accuracy of trajectory correction through interactive operation and coordinate adjustment. The method effectively solves the deficiencies of traditional technologies in data fusion, trajectory completion and visual editing, and provides technical support for vehicle trajectory reconstruction.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, in particular to a vehicle trajectory reconstruction and deviation correction method and device based on spatio-temporal data fusion. BACKGROUND

[0002] The existing vehicle trajectory reconstruction and deviation correction method has obvious deficiencies. The traditional system performs poorly in data acquisition and fusion, and fails to effectively integrate multi-source spatio-temporal information, affecting the trajectory reconstruction accuracy.

[0003] In addition, the existing technology has bottlenecks in trajectory breakage processing and coordinate conversion. Most systems lack perfect forward prediction mechanism and backward tracing strategy, resulting in inaccurate trajectory completion.

[0004] The existing system has technical shortcomings in visual editing. Lack of in-depth use of geographic information, it is difficult to achieve accurate trajectory adjustment through interactive operation, affecting the practicability of the system. The solution of these problems has important significance for improving the trajectory reconstruction effect. SUMMARY

[0005] In view of the problems in the prior art, the present application provides a vehicle trajectory reconstruction and deviation correction method and device based on spatio-temporal data fusion, which can effectively solve the deficiencies of traditional technology in data fusion, trajectory completion and visual editing, and provide technical support for vehicle trajectory reconstruction.

[0006] In order to solve at least one of the above problems, the present application provides the following technical scheme:

[0007] In a first aspect, the present application provides a vehicle trajectory reconstruction and deviation correction method based on spatio-temporal data fusion, comprising:

[0008] Collecting traffic monitoring data, dynamic road network topology data and traffic data, the traffic monitoring data containing camera snapshot timestamp, target vehicle license plate information, position coordinates and coordinate missing node marker, the dynamic road network topology data containing road grade, ramp structure parameters and single lane restriction rules, the traffic data containing floating car trajectory data and real-time traffic data, selecting a reference control point to establish a coordinate conversion matrix, converting the camera position coordinates to geographic information road network coordinate system, and calibrating the timestamps of the traffic monitoring data and traffic data based on global positioning system standard time;

[0009] For the trajectory broken section, the position coordinates, time stamp, and speed information of the last known node before the break are used as the reference, the road speed limit in the dynamic road network topology data and the real-time traffic data are combined to calculate the forward prediction trajectory point, the first discovered node after the break is used as the starting point, the driving characteristics in the probe vehicle trajectory data and the one-way restriction rules are combined to calculate the reverse tracing trajectory point, the forward prediction trajectory point and the reverse tracing trajectory point are weighted and averaged to generate trajectory broken section completion data;

[0010] The trajectory broken section completion data is superimposed on the geographic information road network layer to construct a trajectory editing interface, the node coordinates in the trajectory broken section completion data are read, the trajectory line is drawn in the geographic information road network layer, the positions of the node coordinates are marked, and the node coordinates are adjusted through a drag operation.

[0011] Further, it also includes: obtaining traffic monitoring data stream from the city traffic monitoring platform, parsing the device identification code, snapshot timestamp, license plate number, latitude and longitude coordinate field in the data stream, marking the data node with missing latitude and longitude coordinates as invalid node, extracting the road grade classification table, ramp curvature parameter table, one-way traffic rule table from the road network management system, constructing the dynamic road network topology data structure, and establishing the index mapping of road network nodes and connection relationship;

[0012] Collecting latitude and longitude, driving speed, and timestamp data uploaded by the vehicle positioning device from the probe vehicle data platform, storing the data according to vehicle number, obtaining real-time traffic state data of road grid from the map service platform, extracting average speed data within the road section, associating and saving the average speed data with road grid number, and establishing real-time traffic query index.

[0013] Further, it also includes: obtaining landmark building control point coordinates from the city geographic information database, calculating the relative position relationship of the control points in the local coordinate system and the geographic information road network coordinate system, constructing an affine transformation matrix, calculating the coordinate system conversion parameters according to the affine transformation matrix, applying the conversion parameters to the coordinate system conversion of the camera position coordinates, and generating the camera distribution data in the geographic information road network coordinate system;

[0014] Reading the standard time information of the global positioning system timing center, using the standard time information as the reference time, calculating the deviation values of the camera device timestamp in the traffic monitoring data and the vehicle-mounted device timestamp in the probe vehicle trajectory data relative to the reference time, correcting the timestamp data according to the deviation values, and establishing a unified time reference.

[0015] Further, the method further comprises: extracting the position coordinates, the time stamp and the speed information of the last known node before the trajectory break; searching for the road grid number corresponding to the position coordinates in the dynamic road network topology data; obtaining the road grade speed limit value of the road grid; reading the current average driving speed of the road grid from the real-time road condition data; and performing weighted calculation on the average driving speed and the speed limit value to obtain a predicted reference speed.

[0016] Based on the predicted reference speed, a uniform speed kinematic model is constructed, the position coordinates are substituted into the uniform speed kinematic model, the coordinate position at the next predicted time is calculated, the predicted coordinates are corrected according to the ramp curvature parameters of the road grid, the forward predicted trajectory point is generated, and the prediction calculation is repeatedly performed until the middle time of the broken section is reached.

[0017] Further, the method further comprises: screening trajectory data of the same road section passed by the target vehicle at the same period from the floating car trajectory data; extracting lane preference features and speed variation rules in the trajectory data; reading the position coordinates and the time stamp of the first discovered node after the break; determining a feasible reverse tracing path in combination with the one-way street restriction rule; constructing a reverse trajectory prediction model based on the lane preference features and the speed variation rules; and calculating the reverse tracing trajectory point.

[0018] The forward predicted trajectory point and the reverse tracing trajectory point are paired according to the time stamp, the spatial distance of each pair of trajectory points is calculated, the fusion weight coefficient is set according to the spatial distance, the paired trajectory points are weighted and averaged, the fusion trajectory point is generated, and the fusion trajectory point is connected in time sequence to construct the trajectory broken section completion data.

[0019] Further, the method further comprises: loading vector data of geographic information road network; constructing a road network rendering layer; drawing road grade, ramp position, one-way identification and other topological elements on the road network rendering layer; reading the node coordinate sequence in the trajectory broken section completion data; creating a trajectory drawing layer above the road network rendering layer; converting the node coordinate sequence into a Bezier curve control point to generate a smooth trajectory line.

[0020] Based on the grid layout, a trajectory editing canvas is created, the road network rendering layer and the trajectory drawing layer are mapped to the coordinate system of the editing canvas, the position coordinates of the trajectory nodes are marked, a node dragging controller is constructed in the editing canvas, the node dragging controller is associated with the trajectory node position, and a coordinate transformation response mechanism is established.

[0021] Further, the method further comprises: extracting a node coordinate sequence from the trajectory broken section completion data, calculating a direction vector and a distance between adjacent nodes, generating trajectory line control points based on a third-order Bezier curve interpolation algorithm, connecting the control points into a smooth trajectory line, marking node positions in a geographic information road network layer, drawing node icons, and establishing a mapping relationship between the node icons and the coordinate data;

[0022] An interactive layer is constructed, a node dragging response area is created in the interactive layer, a mouse press event is listened to, a real-time coordinate change amount in a dragging process is obtained, a node position is updated according to the coordinate change amount, trajectory line control points are recalculated, a trajectory line shape is dynamically updated, and adjusted node coordinate data is recorded.

[0023] In a second aspect, the application provides a vehicle trajectory reconstruction and deviation correction device based on spatiotemporal data fusion, comprising:

[0024] A data processing module is configured to collect traffic monitoring data, dynamic road network topology data, and traffic data, wherein the traffic monitoring data includes a camera snapshot timestamp, target vehicle license plate information, position coordinates, and a coordinate missing node marker, the dynamic road network topology data includes road grades, ramp structure parameters, and single-lane restriction rules, and the traffic data includes floating car trajectory data and real-time traffic data; a coordinate conversion matrix is established by selecting a reference control point, the camera position coordinates are converted to a geographic information road network coordinate system, and the timestamps of the traffic monitoring data and the traffic data are calibrated based on a global positioning system standard time;

[0025] A trajectory completion module is configured to, for a trajectory broken section, take the position coordinates, timestamps, and speed information of the last known node before the break as a reference, calculate forward prediction trajectory points in combination with road speed limits in the dynamic road network topology data and real-time traffic data, take the first discovered node after the break as a starting point, calculate reverse tracing trajectory points in combination with driving features in the floating car trajectory data and single-lane restriction rules, and generate trajectory broken section completion data by performing weighted average fusion on the forward prediction trajectory points and the reverse tracing trajectory points.

[0026] A reconstruction and deviation correction module is configured to superimpose the trajectory broken section completion data on a geographic information road network layer, construct a trajectory editing interface, read node coordinates in the trajectory broken section completion data, draw a trajectory line in the geographic information road network layer, mark the positions of the node coordinates, and support adjustment of the node coordinates through a dragging operation.

[0027] In a third aspect, the application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the vehicle trajectory reconstruction and deviation correction method based on spatiotemporal data fusion when executing the program.

[0028] In a fourth aspect, the present application provides a computer readable storage medium, having stored thereon a computer program, which, when executed by a processor, implements the steps of the vehicle trajectory reconstruction and rectification method based on spatio-temporal data fusion.

[0029] In a fifth aspect, the present application provides a computer program product, comprising computer programs / instructions, which, when executed by a processor, implement the steps of the vehicle trajectory reconstruction and rectification method based on spatio-temporal data fusion.

[0030] From the above technical solutions, the present application provides a vehicle trajectory reconstruction and rectification method and device based on spatio-temporal data fusion. Through the innovative design of a multi-source data fusion model, the unified processing of data is achieved through coordinate conversion and time calibration. A trajectory completion system is constructed, and a reliable trajectory reconstruction mechanism is established in combination with road network characteristics and real-time road conditions. Visualization editing is introduced, and the accuracy of trajectory correction is ensured through interactive operation and coordinate adjustment. This method effectively solves the deficiencies of traditional technologies in data fusion, trajectory completion, and visualization editing, and provides technical support for vehicle trajectory reconstruction. BRIEF DESCRIPTION OF DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0032] Figure 1 A flowchart of the vehicle trajectory reconstruction and rectification method based on spatio-temporal data fusion in the embodiments of the present application;

[0033] Figure 2 A structural diagram of the vehicle trajectory reconstruction and rectification device based on spatio-temporal data fusion in the embodiments of the present application;

[0034] Figure 3 A structural diagram of the electronic device in the embodiments of the present application.

[0035] Reference signs:

[0036] Electronic device 9600, central processor 9100, memory 9140, communication module 9110, input unit 9120, audio processor 9130, display 9160, power supply 9170, buffer memory 9141, application / function storage unit 9142, data storage unit 9143, driver program storage unit 9144, antenna 9111, speaker 9131, microphone 9132. DETAILED DESCRIPTION

[0037] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0038] The acquisition, storage, use and processing of data in the technical solutions of the present application all comply with relevant provisions of laws and regulations.

[0039] In view of the problems in the prior art, the present application provides a vehicle trajectory reconstruction and deviation correction method and device based on spatio-temporal data fusion. By innovatively designing a multi-source data fusion model, the method realizes unified processing of data through coordinate conversion and time calibration. A trajectory completion system is constructed, and a reliable trajectory reconstruction mechanism is established in combination with road network characteristics and real-time road conditions. Visualization editing is introduced, and the accuracy of trajectory correction is ensured through interactive operation and coordinate adjustment. The method effectively solves the deficiencies of traditional technologies in data fusion, trajectory completion and visualization editing, and provides technical support for vehicle trajectory reconstruction.

[0040] In order to effectively solve the deficiencies of traditional technologies in data fusion, trajectory completion and visualization editing, and provide technical support for vehicle trajectory reconstruction, an embodiment of a vehicle trajectory reconstruction and deviation correction method based on spatio-temporal data fusion is provided in the present application, as shown in Figure 1 The vehicle trajectory reconstruction and deviation correction method based on spatio-temporal data fusion specifically includes the following contents:

[0041] Step S101: Collect traffic monitoring data, dynamic road network topology data and traffic data. The traffic monitoring data includes a camera snapshot timestamp, target vehicle license plate information, position coordinates and coordinate missing node markers. The dynamic road network topology data includes road grades, ramp structure parameters and single lane restriction rules. The traffic data includes floating car trajectory data and real-time road condition data. A coordinate conversion matrix is established by selecting a reference control point, the camera position coordinates are converted to geographic information road network coordinates, and the timestamps of the traffic monitoring data and the traffic data are calibrated based on the global positioning system standard time.

[0042] Optionally, the embodiment falls in the entrance stage S101 of the system, and the engineering team lays a solid foundation around the three things of data collection, coordinate unification and time calibration. We first pull through the three data sources: subscribe to the card data stream from the urban traffic monitoring platform, parse the device identifier, snapshot timestamp, license plate number and latitude and longitude fields; if the latitude and longitude is found to be empty or out of bounds, immediately mount the "coordinate missing marker" on the node structure, and keep the original timestamp and device ID, which can be filled back by road network matching or operation and maintenance in the future. Parallel threads pull the road level classification table, ramp structure parameters and one-way traffic rules from the road network management system, forming a weighted directed graph, with road geometry nodes and card anchor points as nodes, and speed limit, lane number, curvature radius and single lane as edge weights. On the floating car side, we gather latitude and longitude, speed and timestamp by vehicle ID and time sequence, and bind the real-time speed of the road grid of the traffic platform with the grid number as a key value, and write it into an index that can be retrieved by "grid ID + time period". In this way, the three types of data have their own sources, time periods and spatial indexes when they reach the cache pool, and subsequent fusion will not be a tangled mess.

[0043] The embodiment does not assume in coordinate unification, but uses control points to pull the card coordinates from the local system of each device back to the geographic information road network coordinate system. We select multiple sets of control points recognized by the municipal department, and the control points have high-precision coordinates in the local coordinate system and the road network coordinate system, and the number is not less than three sets and covers the area as much as possible. The program calculates the relative position relationship of these same points, estimates the scale, rotation and translation, constructs an affine transformation matrix, and then uses RANSAC-style robust fitting to remove large points with linear residual anomalies, avoiding individual camera installation errors from distorting the whole. After the matrix is stable, the card position coordinates are converted to the road network coordinate system in batches, and a drop plot verification is immediately performed after conversion: map the card to the nearest road centerline, and if it deviates more than the tolerance threshold, it is recorded as "suspected misplacement" and will be reviewed again later using ramp structure and traffic rules. The selection and robust fitting of control points is one of the key patent points here, and the reason is clear - only when the space is unified reliably, can the trajectory inference and fracture completion not be based on the wrong graph.

[0044] This embodiment uses a unified time reference to align the clock differences of each device in time. We obtain a continuous standard time stream from the GPS time center as the time axis zero point. For each set of floating car devices and each set of loop devices, we calculate the deviation of their timestamps from the standard time and distinguish between the "fixed deviation" (the overall lag or lead of the internal clock) and the "drift component" (the slow change caused by temperature drift or slight power fluctuations). The estimation method is very simple: in a short time window, use the last N samples to do linear fitting, and the intercept corresponds to the fixed deviation, and the slope corresponds to the drift rate. When correcting, the fixed deviation is applied first, and then the drift rate is used to slightly stretch or compress the different sampling times to generate the calibrated unified timestamps. Loop and floating car have different coverage densities, so the sampling window and N will be configured separately to ensure that neither the noise is amplified nor the noise is smoothed. The significance of time calibration is not only to align the recording order, but also directly related to whether the "forward prediction / backward tracing" meeting point can meet in the correct time slice.

[0045] After straightening the space and time lines, this embodiment begins to associate and record data. Loop points are written into a spatial index tree with the road network coordinate system as the reference, and floating car trajectories are subjected to light map matching to hang each trajectory point to the corresponding road grid and the nearest road segment ID. The average speed data of real-time traffic is stored in columnar storage with "grid ID + time slice" as the key, and the spatiotemporal index allows subsequent second-level retrieval of its corresponding speed background near the broken section. For loop nodes with missing coordinates, do not delete them first, mark them as "to be positioned", keep the license plate and timestamp, and then use the time difference between the nearest neighbor and the adjacent effective loop in the road network to infer their probable position; this "soft deletion" strategy is crucial in security case review, because some devices lose GPS lock for a short time, but the license plate information is still reliable and should not be discarded.

[0046] This embodiment supplements an equation for self-checking before entering the warehouse on the engineering side, which controls the fault tolerance threshold of different data sources: E = Es + Et. In the equation, E represents the comprehensive error estimate value of a single record; Es is the spatial error, which comes from the fitting residual of the control point and the projection distance of the point to the road center line; Et is the time error, which comes from the combination of the fixed deviation and the drift fitting residual of the device. We set E to be less than the threshold to be included in the "highly reliable library", and the records exceeding the threshold enter the "to be reviewed library". The parameter meaning is very intuitive: Es measures whether it is on the correct road or not, and Et measures whether the clock is reliable or not. Subsequent trajectory completion prefers to take the highly reliable library to avoid low-quality samples from biasing the model.

[0047] This example talks about two field variants, which are the key points in the real environment. One is the dense area of elevated ramp, the curvature of ramp and the single lane constraint are complex, and the control point density is not enough. Affine will appear local stretching, we increase the two-stage refinement in this area: large range uses affine, small range does thin plate spline interpolation on the road skeleton, and local nonlinearity is included; Another is the old card hole device, which uses local NTP to correct time but the network jumps are large, and Et is high for a long time. We enable segmented fitting in time correction, introduce breakpoints when encountering jump segments, re-estimate fixed deviation, and avoid thinning the instantaneous jump error to the whole segment data.

[0048] This example falls back to the method target, S101 truly solves the three basic problems of "coordinate inconsistency, time misalignment, and data quality difference". Once this step is solid, the subsequent S201-S205 forward prediction and reverse tracing of the broken segment can be calculated in a unified space-time grid, the speed limit, ramp curvature, and single lane rule can correctly constrain the candidate path, and the speed background of the floating car can be referenced in the accurate time slice. Engineers can open the timeline of a certain car on the interface, see the card hole point distribution close to the road line, and there is no "fast and slow clock" of discrete devices on the time axis, which is more psychologically stable and more controllable in algorithm.

[0049] Step S102: For the trajectory broken segment, taking the position coordinates, timestamp, and speed information of the last known node before the break as the reference, combining the road speed limit in the dynamic road network topology data and the real-time traffic data to calculate the forward prediction trajectory point, taking the first discovered node after the break as the starting point, combining the driving features in the floating car trajectory data and the single lane restriction rules to calculate the reverse tracing trajectory point, and performing weighted average fusion on the forward prediction trajectory point and the reverse tracing trajectory point to generate trajectory broken segment completion data;

[0050] Optionally, this example falls on the core bridging step S102 of trajectory reconstruction, the scene is a road section where urban trunk roads and ramps are intertwined, and there are intermittent coordinate missing in the monitoring card hole. We first take the longitude and latitude, timestamp t , and instantaneous speed v of the last known node N before the break as the starting condition, locate its grid and adjacent reachable edges in the dynamic road network topology, read the road grade speed limit, ramp curvature, and one-way rule, and then get the average driving speed of the grid and the next hop from the real-time traffic index. Here is a judgment: if v Rather than blindly trust either party, a predicted baseline speed v b is calculated, with recent traffic congestion given more weight. A forward-evolution model is then constructed, which uses v b to extrapolate along candidate edges hour by hour, with ramp access modified according to curvature parameters, and maximum turning angles limited, until the time rolls to the break midpoint or approaches the time t + of the first discovered node N + after the break. In this traversal, each time step not only records the predicted point coordinates, but also records the confidence of the selected edge, and the weight traces of the speed sources, in order to give the subsequent fusion an interpretable "physical credentials".

[0051] The embodiment takes a "backwards from N + " strategy in reverse tracing. We read the coordinates and timestamp t + of N +, and enumerate the upstream edges that can be reversely accessed under the one-way rule constraints of the road where N + is located. Topology alone is not enough, we filter out from the probe vehicle trajectory data a fleet sample that is close in time to the target vehicle, and passes through the same road segment, and extract two types of driving features: one is lane preference (for example, at a straight-right turn junction with multiple lanes, vehicles on the right lane are more inclined to choose ramps), and the other is speed change pattern (deceleration-turning-acceleration time profile). These features are naturally consistent with human driving habits, and the closer the values are to the congestion or free-flow situation, the more credible the reverse model. A lightweight reverse trajectory predictor is thus established to "pull back" N +, at each time step according to the one-way restrictions to eliminate illegal edges, and then according to lane preference and speed pattern scoring to select the most likely upstream edge for backtracking, until the time rolls back to the break midpoint or reaches the time t + around the vicinity. The reverse sequence and the forward sequence overlap in the time axis, and there may be a spatial misalignment between the two, which is the material for the next fusion.

[0052] The embodiment will complete time pairing before weighted average fusion. We take a uniform time granularity Δt in [t ​We establish sampling points on [t+], and interpolate the forward prediction set F(t) and the backward tracing set R(t) to ensure that there is a pair of candidate points at each sampling time. The spatial distance d(t) between each pair of points is an intuitive measure of the consistency between the two pieces of evidence. We set a set of weights related to d(t), local topological confidence c_topo(t), and traffic flow consistency c_flow(t), and then calculate the fused coordinates. Here, a single formula is given to illustrate the weight allocation logic: w_f(t) = 1 / (1 + α·d(t)) · c_topo_f(t), w_r(t) = 1 / (1 + β·d(t)) · c_flow_r(t), which are normalized to obtain the final weights W_f and W_r. Here, d(t) is the spherical distance between the forward and reverse points, reflecting the spatial conflict between the two pieces of evidence; α and β are distance sensitivity coefficients, representing the degree of penalty for the conflict; c_topo_f(t) comes from the score of the forward path on topological legality, including speed limit compliance and the rationality of ramp angles; c_flow_r(t) comes from the similarity between the reverse sample and the real-time traffic flow speed profile. A weighted average of W_f and W_r is used to generate the fused trajectory point G(t). The physical meaning of the parameters falls within the realm of traffic and geometric common sense, and does not violate natural laws: the closer the distance, the more compliant the topology, and the more consistent the traffic flow, the higher the weight.

[0053] This embodiment performs two constraint filters after fusion. The first is a road alignment method, which fine-tunes G(t) according to the centerline of the nearest feasible lane. If the deviation exceeds a threshold, the candidate edge set at the corresponding time step is reviewed, and the edge that better matches the speed limit and curvature is prioritized for correction. The second is a speed continuity constraint, which calculates the adjacent speeds and accelerations of the G(t) sequence. If an unreasonable spike occurs (e.g., a momentary reversal on a straight one-way road), the time step where the spike occurred is backtracked to check whether the forward evidence or the reverse evidence is weak, and the corresponding weights are dynamically reduced and the fusion is performed again. After these two constraints, the broken segment completion data is obtained, including time-series node coordinates, the aligned road ID, and a confidence curve prepared for manual verification.

[0054] This embodiment considers two typical variants to illustrate adaptability. In areas with dense elevated ramps, the joint constraints of curvature and speed limits are stronger, and the forward model is particularly sensitive to "when to enter the ramp." In this scenario, we increase the weight of c_topo to reduce the offset caused by occasional detours from back samples. In old urban areas with dense one-way networks and sparse floating cars, when back samples are insufficient, c_flow will automatically shrink, and the fusion weights rely more on the forward and topology, which is more stable than simply filling in the gaps. Additionally, rain and snow cause widespread speed reductions, and real-time traffic conditions lower v_rt, naturally decreasing v_b. The extrapolation step size is shortened, which can prevent cars from being "over-pushed."

[0055] The technical problems solved by the embodiment are concentrated in three points: first, the time-space fracture caused by the missing bayonet coordinate, the one-way topology and the ramp geometry make the naive straight line interpolation unreliable; second, the coexistence of noise in multi-source data, both device time synchronization error and floating car sampling sparsity; third, the cost of manual correction is too high. Through "forward physical extrapolation + backward group behavior backtracking + adaptive weight based on conflict", the fracture section is filled into an interpretable curve, and each time step leaves a trace of origin. In the subsequent trajectory editing interface, manual correction is required only by dragging a few nodes, rather than starting from scratch. Combining S102 with the time reference calibration and coordinate conversion of the previous sequence, all inputs are pulled back to the same coordinate system and clock, and the basis for determining the fusion weight is also solid. When the engineer reviews it again, he can see the origin and development of each reasoning from the log.

[0056] Step S103: superimpose the trajectory fracture section completion data onto the geographic information road network layer to construct a trajectory editing interface, read the node coordinates in the trajectory fracture section completion data, draw a trajectory line in the geographic information road network layer, mark the positions of the node coordinates, and support adjusting the node coordinates through a dragging operation.

[0057] Optionally, for step S103, the fracture section completion data output by the previous step S102 is put on the interactive map interface for verification and repair. We first load the vectorized geographic information road network into memory to construct a road rendering layer and an interactive index. The index key consists of a three-tuple of "road ID-direction-lane number" to facilitate subsequent binding of the trajectory to a specific lane. Before the trajectory fracture section completion data is entered, a light check is performed: check if the node time sequence is monotonic, the coordinates are within the city boundary range, and there is no single-row conflict with the corresponding road. After passing the check, the node sequence is sent to the trajectory drawing pipeline, which consists of three steps: first, project the node coordinates to the rendering coordinate system, and keep the original geographic coordinates for rewriting; second, generate third-order Bezier or Catmull-Rom control points based on the direction and curvature of adjacent nodes; third, draw the trajectory line adaptively sampled by resolution in the rendering thread to avoid jagged or excessive smoothing in dense node sections.

[0058] The embodiment in the "superimposed" link is not simply covered, but in the way of layer synthesis to put the track on the road. Road layer is responsible for the performance of road level, ramp entrance, one-way arrow, track layer carries two types of elements: line elements (fusion track line) and point elements (key nodes). In order to reduce visual congestion, we introduce the symbolization rule based on speed and curvature: straight line segment with thin line, curved segment with slightly thick line and light color direction scale; Key nodes are colored according to the confidence output of S102, and the deeper the color, the more certain the model, and the user can see at a glance which points need to be checked. Point elements and their geographic coordinates establish a bidirectional mapping, and any rendering coordinate transformation (scaling, rotation) will be synchronized to the screen position of the point through matrix multiplication, while the logical coordinates always remain in the geographic coordinate system, avoiding the accumulation of repeated projection errors.

[0059] The embodiment puts the editing ability on the "node drag controller". The controller opens a hot zone for each node in the interaction layer, listening to the press, move, and release events. When pressed, record the original geographic coordinates, the road ID and the adjacent node index; During the drag process, the screen displacement is inversely transformed into geographic coordinate increment in real time, and the new candidate coordinates are obtained, and then the "road fitting method" is used for constraint: preferentially adsorb the node to the nearest drivable lane center line, if it conflicts with the one-way direction, give a light prompt and limit the adsorption to the nearest edge that meets the direction. When released, trigger the incremental update, recalculate the control points of the adjacent two segments of the node, and update the trajectory curve locally, rather than redraw the whole line, which is more stable in rendering efficiency. In order to avoid unreasonable position of the user to drag the point, we set the speed continuity check, calculate the speed and acceleration of the two segments before and after the node after the drag, if the transition is too large, it will be rolled back to the nearest feasible point within the boundary, and the user interface only prompts "converged according to road and dynamics constraints".

[0060] The embodiment maintains a "retraceable editing log" at the data level. Each drag records four pieces of information: node ID, original coordinates, new coordinates, and triggered constraint type (road, one-way, speed). The log is stacked in chronological order, supporting step-by-step undo and redo. More importantly, the log will bring the "source label" of the node, that is, whether the point comes from "forward prediction" or "backward tracing" in S102, or the weight proportion of the fusion of the two. Engineers judge accordingly, if a point originally has low weight, the manual adjustment amplitude can be more bold; If the weight is high, it is more cautious. This "evidence-based visual and editable" is the core point of the application in the cooperation of man and machine, avoiding the embarrassment of black box output difficult to correct.

[0061] This embodiment takes into account the difficulty of human-computer interaction at complex intersections, and extends two convenient tools. One is "segment-level dragging", after the user selects a segment of the trajectory, the midpoint of the segment is used as the anchor for overall translation or micro-rotation, the system distributes the transformation to each node in the segment, keeps the relative arc length and curvature distribution approximately unchanged, and is suitable for processing overall offset. The other is "road replacement", at the multi-layer road surface superposition under the overpass, the user selects the target road layer, the system immediately remaps the road ID and center line adsorption for the segment nodes, automatically corrects the problem of misbinding of elevated roads and ground, and reduces the physical work of point-by-point correction.

[0062] This embodiment is to let the engineering personnel have a consistent experience on the large screen and the mobile terminal, and the rendering and interaction logic adopts resolution adaptation. When zooming to the city level, the nodes are aggregated into sparse sampling points, and dragging is temporarily closed to prevent misoperation; only when zoomed to the block level, fine-grained interaction is opened. Each time the view changes, the control point sampling step is re-evaluated to ensure smooth lines without occupying too much GPU time. The ranging and time scale tools in the interface are also linked to the trajectory, and the user can see the time span and average speed of the segment by sliding on the segment, which is convenient for judging whether further adjustment is needed without switching panels.

[0063] The technical effects of this embodiment are embodied in three aspects. First, the broken segment completion data is presented in a regularized layer system, and the user can clearly distinguish the "determination and uncertainty" of the model output, and prefer to focus on the dark low-confidence nodes. Second, the dragging action is constrained by the road, one-way and dynamics, and it is not easy to drag the trajectory out of the "common sense boundary", and the correction process is stable and can be traced back. Third, local update and segment-level tools reduce the interaction cost, and even in the face of long breaks, it can converge to a reliable curve in a limited time. Looking at S103 together with the time and space unification of the previous S101 and the bidirectional prediction of S102, the closed loop of "look, change and prove" of the data is complete, and the subsequent export and case research report also have a basis, and it is no longer a line drawn by feeling.

[0064] From the above description, the vehicle trajectory reconstruction and correction method based on spatio-temporal data fusion provided by the embodiments of the present application can realize unified processing of data through innovative design of a multi-source data fusion model, through coordinate conversion and time calibration. A trajectory completion system is constructed, and a reliable trajectory reconstruction mechanism is established in combination with road network characteristics and real-time road conditions. Visual editing is introduced, and the accuracy of trajectory correction is ensured through interactive operation and coordinate adjustment. The method effectively solves the deficiencies of traditional technologies in data fusion, trajectory completion and visual editing, and provides technical support for vehicle trajectory reconstruction.

[0065] In an embodiment of the vehicle trajectory reconstruction and correction method based on spatio-temporal data fusion of the present application, the following contents can also be specifically included:

[0066] Step S201: Obtain traffic monitoring data stream from the urban traffic monitoring platform, parse the device identification code, snapshot timestamp, license plate number, latitude and longitude coordinate field in the data stream, mark the data node with missing latitude and longitude coordinates as invalid node, extract the road grade classification table, ramp curvature parameter table, one-way traffic rule table from the road network management system, construct the dynamic road network topology data structure, and establish the index mapping of road network nodes and connection relationship;

[0067] Step S202: Collect latitude and longitude, speed, and timestamp data uploaded by the vehicle positioning device from the floating car data platform, group and store the data according to the vehicle number, obtain real-time traffic state data of the road grid from the map service platform, extract average speed data within the road section, associate and save the average speed data with the road grid number, and establish a real-time traffic query index.

[0068] Optionally, the embodiment is positioned in the construction links S201 and S202 of the data base, and the goal is to integrate the monitoring camera, road network topology, and floating car and road condition flow into a queryable and traceable structure. We first open a stable message channel for the urban traffic monitoring platform in the access layer, and parse the device identification code, snapshot timestamp, license plate number, and latitude and longitude coordinates of the data stream one by one. Parsing is not only field unpacking, but also legal verification: latitude and longitude out of bounds, zero coordinates, reverse timestamp, unknown device ID, etc. are marked immediately. Especially for records with empty or invalid latitude and longitude, mark them with "missing coordinates" and put them into the invalid node set, but do not delete the license plate and timestamp, because these clues may be needed for subsequent map matching or manual review. At the same time, the system pulls three types of structured data, road grade classification table, ramp curvature parameter, and one-way traffic rule, from the road network management system, cleans up historical closed roads and expired rules, and extracts road geometric nodes as graph vertices, constructs directed edges according to the driving direction, and the edge weight contains the grade speed limit, lane number, curvature radius, and prohibited turning sign. On this dynamic graph, establish a double-layer index: one layer is the spatial R-tree or grid index, which supports "given latitude and longitude, find the nearest road section and topology adjacency"; the other layer is the topology adjacency table, which maps to the upstream and downstream connection relationship according to the road section ID, which is convenient for subsequent forward / backward search. After the construction is completed, the camera anchor point and the road node establish a reference relationship, and a camera record can be found in O(logN) time in the future.

[0069] The embodiment is not lucky in data quality, and each card hole record parsed will try to make a rough match with the road network node to see if the latitude and longitude projection to the nearest road center line is within the tolerance. If it is out of range but the coordinates are not missing, it is classified as "suspected misplacement" and added to the pending library. The reason for doing this is that the subsequent trajectory reconstruction is extremely dependent on the correct binding of "card hole-road", and early release will bring errors into the prediction model of S102. Topological data needs to be updated regularly. City road construction changes frequently. The system compares the differences between the old and new versions in the night batch task, updates the edge weight and one-way attribute, and refreshes the adjacency index to ensure that the calculation of the next day is consistent with the rules of the day.

[0070] The embodiment enters S202 to integrate the probe car and traffic flow. The latitude, longitude, speed, and time stamp uploaded by the vehicle positioning device are decoded first, then grouped by vehicle number, and then sorted by time within the group. Remove duplicates and obvious jump points (based on unreasonable speed threshold and acceleration mutation rules), leaving high-quality trajectory segments. Each trajectory point tries to connect to the road network: according to its latitude and longitude, use spatial indexing to find candidate road segments, consider one-way restrictions and the driving direction of the vehicle's two points before and after, and select the most reasonable one as the road segment ID of the point. This step generates a "vehicle ID-time-road segment" triple, which prepares for subsequent speed profile statistics. On the traffic side, we pull the real-time traffic state of the road grid from the map service platform by time slice, parse the interval average speed and bind it to the grid number into the database, build a "grid ID+time slice→average speed" query index, and additionally record the data source time and confidence level to avoid misuse of old data in new time periods.

[0071] After connecting the two data, the embodiment starts the cross-source linkage. For card hole records, if the latitude and longitude are valid, they will fall into the nearest road under spatial indexing, and the average speed of the grid where the road is located in the record time slice will be read as the background; if the card hole coordinates are missing, but the same vehicle before and after has a trajectory passing through the adjacent road segment in the probe car library, we temporarily locate this card hole record to the adjacent position of the trajectory, mark it as "inferred coordinates", and do not participate in hard constraints, only for manual or subsequent model reference. Conversely, rare paths in the probe car trajectory (such as crossing unpopular ramps) will be recorded as "sample fingerprints" and added to the confidence field of adjacent edges in the topology graph, making the subsequent S102 backtracking more "understandable".

[0072] The embodiment gives a logic judgment before warehousing, and the purpose is to convert the multi-source consistency into an executable threshold: M = I_geo ^ I_time ^ I_topo. In the formula, M is a Boolean judgment of whether to enter the "highly trusted library"; I_geo represents whether the spatial consistency is met, specifically whether the projection distance of the lens or track point to the target road center line is less than the set threshold; I_time represents whether the time consistency is met, specifically whether the recording time is covered in the effective window of the road condition time slice and no device clock jump occurs; I_topo represents whether the topological consistency is met, specifically whether the single-row, no-turn, ramp entrance direction conflicts with the point-to-point calculated direction. The physical meaning of the three quantities is clear: the space must fall on the road, the time must be in the correct slice, and the topology must not violate the rules. Only when M is true, the corresponding record enters the highly trusted index, otherwise it is marked into the pending review area.

[0073] The embodiment considers two field variants to illustrate scalability. One is the multi-layer road area under the subway cover or viaduct, and the GPS multipath effect is serious. The floating car point cloud presents "scattered clusters". We introduce short window trajectory smoothing before connecting the road section, and let the I_geo threshold adaptively adjust with the road grade: the threshold is wider on the highway, and the threshold is tighter on the branch, so as not to discard the high-speed error too early. The second is that the road condition platform is not updated for a short time due to an emergency, and we maintain the data "time limit water level" for each grid in the query index. Once it falls below the warning line, I_time is directly false, but it does not affect the trajectory warehousing. Only the average speed of these grids is prohibited from being used as a prediction basis to prevent distorted background speed from being brought into the fusion.

[0074] The technical effect of the embodiment is reflected in three lines being tightened: the monitoring flow is sorted from "raw records" into "road anchor points with topological constraints", the floating car is changed from "point set" to "trajectory on the road", and the road condition is deposited from "platform snapshot" to "speed field that can be retrieved by grid and time slice". For subsequent trajectory reconstruction, forward prediction can directly read the road speed limit and real-time speed, backward tracing can capture driving habits from sample fingerprints, and artificial review can visually determine the data trust level on the interface. In this way, S201 / S202 not only scans the data into the library, but also prepares the constraints, confidence and index required by subsequent algorithms. The subsequent calculation does not have to be distracted by the "inconsistency and instability" of the basic data.

[0075] In an embodiment of the vehicle trajectory reconstruction and deviation correction method based on spatio-temporal data fusion provided in the application, the following contents can be specifically included:

[0076] Step S301: Obtain landmark building control point coordinates from the urban geographic information database, calculate the relative position relationship of the control points in the local coordinate system and the geographic information road network coordinate system, construct an affine transformation matrix, calculate the coordinate system conversion parameters according to the affine transformation matrix, apply the conversion parameters to the coordinate system conversion of the corner position coordinates, and generate the corner distribution data in the geographic information road network coordinate system;

[0077] Step S302: Read the standard time information of the global positioning system timing center, take the standard time information as the reference time, respectively calculate the deviation values of the corner equipment time stamp in the traffic monitoring data and the vehicle-mounted equipment time stamp in the floating car trajectory data relative to the reference time, correct the time stamp data according to the deviation values, and establish a unified time reference.

[0078] Optionally, the present embodiment is developed around steps S301 and S302, and the scenario is a stock city with multi-manufacturer corner mixed distribution and local coordinates for each camera. We first retrieve a set of evenly distributed landmark building control points from the urban geographic information database, which covers viaducts, ground and both sides of the river as much as possible. The control points have high-precision coordinates in both the local coordinate system and the geographic information road network coordinate system (GIS road network system). In order to reliably project the corner coordinates from the local system to the GIS road network system, we construct an affine transformation with the same point pair: calculate the scale, rotation and translation parameters, the initial value comes from the least squares closed-form solution, then remove gross error points with a residual threshold, and then iterate on the remaining set to prevent individual camera installation errors from deviating the global system. Considering that there may be slight nonlinear distortion in the urban block, we model by administrative grid: the global affine is used in the large area, and the first-order elastic correction is made under the constraint of the road skeleton in the small area to avoid edge stretching in the sparse control point area. Finally, a mapping table of "region ID→affine matrix" is formed, and the matrix is applied to the corner position coordinates in batches after stabilization to obtain the corner distribution point set in the GIS road network system.

[0079] The present embodiment does not regard coordinate conversion as a one-time arithmetic. Immediately after conversion, "falling road review" is carried out, that is, the projection distance of the corner point to the nearest road center line is calculated. If the distance exceeds the threshold but is still within the road buffer, the nearest ramp curvature and road grade are used to correct the direction, and the point position is adjusted to converge towards the road normal; if the threshold is exceeded and there is a single conflict, the point is marked as "suspected misplacement" and added to the review list. During the review process, the system retrieves the floating car segment that has passed through the corner, compares whether the real passing path and the road ID of the current corner point are consistent, and if they are not consistent and the floating car sample is stable, the road ID consistent with the sample is selected as the priority and the offset is written back. Through this closed loop, the corner anchor point is not only put back into the correct coordinate system, but also truly "stuck" to the road element that can carry subsequent constraints.

[0080] The time reference calibration of the embodiment enters S302, the idea is to take the GPS timing center as the zero point, respectively estimate the time deviation of the lens device and the floating car device. We continuously read the standard time pulse, aggregate its original timestamp according to the device ID, construct a short window linear drift model, and the window length is adaptive according to the device message frequency. Fit two quantities, fixed deviation and drift rate, for each device. The fixed deviation reflects the overall fast and slow clock, and the drift rate reflects the slow change caused by temperature and power supply. Correction, first subtract the fixed deviation from all time stamps, then do linear scaling on the time axis according to the drift rate, and get the calibration time under the unified reference. In order to avoid the "time jump" pollution caused by network jitter, we set breakpoint detection on the time series. When a mutation occurs, the window is cut off, segmented and re-estimated to ensure that the drift only describes the slow change and is not dragged by the mutation.

[0081] The embodiment considers the influence of space and time on the subsequent algorithm together. The lens coordinates are pulled into the GIS road network system, and the forward extrapolation during trajectory reconstruction can correctly read the "speed limit of the road segment and ramp curvature"; the timestamp is pulled to the unified reference, and the time when the forward and reverse sequences meet at the midpoint is meaningful. In order to make the engineering caliber consistent, we do joint verification on each lens record before entering the database: check "whether the projection distance is in line with the rules, whether the single-line rules conflict, and whether the time is aligned with the road condition time slice". If any of them is not satisfied, it enters the pending review area and records the reason. During subsequent S102 calculation, it is defaulted to reduce its weight, so that high-confidence samples can speak first.

[0082] The embodiment provides a discriminant for one-line colleagues to control risk on the engineering side: R = λ·Es + μ·Et. In the formula, R is the comprehensive risk value of the record; Es is the spatial error measure, which takes the projection distance of the lens point to the nearest road center line and the direction penalty; Et is the time error measure, which takes the weighted sum of the fixed deviation residual and the drift fitting residual of the device; λ, μ are weights from business configuration, indicating the emphasis on spatial and temporal credibility in the current city or case. The physical meaning of the parameters is simple: the smaller Es indicates that "the point is more reliable on the road", and the smaller Et indicates that "the clock is more accurate". When R exceeds the threshold, the system does not directly discard, but reduces the weight and enters the database, and generates a review task, ensuring that the data is neither too harsh nor permissive.

[0083] The embodiment also considers the adaptation of two types of special topography. In the stacked area, the control points are often blocked, resulting in sparsity, we introduce "virtual control points" in S301: derive auxiliary points from the bridge piers and ramp stakes of Gaode or local GIS, although the accuracy is slightly lower, but in the affine fitting, participate with small weight, improve local constraints; the streets in old urban areas are narrow and the lens manufacturers are diverse, with large time synchronization differences, in S302 we open a shorter drift estimation window for old devices, and set a more tolerant threshold for Et to avoid excessive correction that disrupts the real traffic flow order.

[0084] The technical effect of the embodiment is more reflected in "availability" and "explainability". After S301, the card hole anchor point is truly on the road, and subsequent speed limit, one-way, and curvature topological constraints have reliable carriers; after S302, all time axes of all sources are compressed to the same scale, and the front and rear evidence of the broken section can be aligned in the correct time slice. The engineer can see the original local coordinates, affine matrix ID, projection distance, risk value R, and time calibration deviation pair of a certain anchor point in the map interface, and if there is a logical trajectory return, it can also be located to which device Et is abnormal or which section of the region Es is too large, so as to decide whether to recalculate, reduce weight, or send a single review.

[0085] In an embodiment of the vehicle trajectory reconstruction and deviation correction method based on spatio-temporal data fusion of the present application, the following contents can also be specifically included:

[0086] Step S401: Extract the position coordinates, timestamp, and speed information of the last known node before the trajectory break, find the road grid number corresponding to the position coordinates in the dynamic road network topology data, obtain the road grade speed limit value of the road grid, read the current average driving speed of the road grid from the real-time road condition data, and calculate the predicted reference speed by weighting the average driving speed and the speed limit value;

[0087] Step S402: Construct a uniform speed kinematics model based on the predicted reference speed, substitute the position coordinates into the uniform speed kinematics model, calculate the coordinate position at the next predicted time, correct the predicted coordinates according to the ramp curvature parameters of the road grid, generate a forward predicted trajectory point, and repeat the prediction calculation until the middle time of the broken section is reached.

[0088] Optionally, the embodiment is developed around steps S401 and S402, and the scene is selected on the road section of the main road connecting ramp, and the card hole is intermittently transmitted in the rainy night, causing the target vehicle trajectory to have a gap. We first read the geographic coordinates p , timestamp t , and instantaneous speed v of the last known node N before the break from the data warehouse with unified spatio-temporal reference S101. In order to avoid single-point speed noise bias, we estimate the stable speed in a short window (such as two or three samples before and after) near N , and combine it with the properties of the road grid where N is located. Through the dynamic road network topology index, p Projecting onto the nearest road segment with the same direction yields the road grid ID g, and two constraints are read: the speed limit V_lim for this grid's road class and the ramp entrance geometry parameters (if they exist). Simultaneously, g is retrieved from the real-time traffic index at time t. The average velocity V_rt in the given time slice. Instead of directly taking either V_rt or V_lim, we construct an interpretable predictive baseline velocity V_b, taking into account both driving physics and traffic conditions: if If V_b is much smaller than V_lim and close to V_rt, it indicates that congestion is dominant, and V_b should be close to V_rt; if If both V_rt and V_lim are high but haven't reached their maximum values, then the vehicle's individual velocity profile is preserved by weighting both. The weights are determined by considering the credibility of two pieces of evidence: the credibility of V_rt increases with the number of samples in the grid and the data's timeliness; and the "hard constraint" of V_lim is used to set an upper bound. As an individual's short-term inertia, it provides a fine-tuning direction.

[0089] In this embodiment, after completing V_b, we proceed to forward extrapolation. We construct a kinematic framework based on the uniform velocity assumption, advancing with a unified time granularity Δt. The first step is to calculate the next time step t. The ideal displacement vector Δs = V_b·Δt·u, where u is the tangential unit vector of the road. The tangential direction is not fixed; it rotates with the arc length parameter of the road centerline. At each step, we find the nearest arc length parameter s on the road centerline to obtain the tangential u(s) and generate the ideal prediction point p' = p + Δs. Considering the steering constraints caused by the curvature of the ramp, we introduce a correction based on the curvature κ(s), making a small correction to p' along the normal vector n(s) to prevent geometric extrapolation from deviating from the lane. If we are about to enter a ramp or curve, κ(s) is larger, and the correction amount increases accordingly; if it is a straight road section, κ(s) approaches zero, and the correction approximately disappears. We also apply two soft constraints to V_b: one is the upper limit of the speed limit to ensure that V_b does not exceed V_lim; the other is the differential damping with V_rt to avoid excessive extrapolation steps in the congested grid, which would cause the vehicle to jump out of the reasonable distance.

[0090] The embodiment does not only look at the nearest center line in candidate path selection, but also uses a topology adjacency table to exhaustively enumerate the next-hop reachable edges. When p' advances to the vicinity of a node, if there are multiple downstream edges (such as the main road straight and the ramp split), we calculate a turning cost for each edge, and the cost is composed of three parts: corner cost (related to the edge incident angle), curvature penalty (high curvature bend is slower), and speed limit difference penalty (the cost of turning from high speed to low speed edge is higher). Take the edge with the minimum cost as the next step of the tangent continuation path, which makes the extrapolation path not easily "jump the ramp", unless the speed and geometry jointly suggest that it should be split. Each forward prediction point generated records the road segment ID, local kappa (s), and the adopted V_b source weight, forming a prediction sequence with a chain of evidence.

[0091] The embodiment encounters the deviation of road network speed and individual speed when engineering lands, and a unique formula is given to explicitly define the calculation logic of V_b: V_b = w_r·V_rt + w_i· , with the constraints w_r + w_i = 1, and 0≤ V_b ≤ V_lim. In the formula, V_rt is the real-time average speed of the road network, reflecting the group traffic state; is the individual stable speed of the vehicle in the short window before the break, reflecting the individual inertia; V_lim is the hard upper bound of the road grade speed limit; w_r and w_i are the group and individual weights respectively, which are jointly determined by the data freshness, sample size, and deviation from V_rt. The physical meaning of this relationship conforms to common sense: the group slows down the individual in congestion, and the individual is closer to the speed limit in smooth traffic; the weights naturally shift with the scene, and there will be no results that contradict the traffic rules.

[0092] The embodiment continues the extrapolation cycle until the middle time t_mid of the broken segment. The reason for stopping at the midpoint is to meet the same time anchor point with the point returned from N+ in the next step. At the end of each step, we also perform two health checks: one is the speed continuity, which calculates the estimated speed and acceleration of the adjacent two points, and eliminates accidental spikes; the second is the road fitting degree, which measures the projection distance of the prediction point to the center line, if it exceeds the threshold, then back half a step, and use a more conservative Δt to recalculate. For the entry time of long radius ramp, we refer to the local gradient change of V_rt and the speed limit difference of adjacent edges. If V_rt has a significant step-down before entering the ramp, the model is more inclined to slow down and split in advance; on the contrary, if there is no obvious change, delay entering the ramp and maintain the main road extension.

[0093] The embodiment gives variants for two special scenarios. The first type is the overlapping place of elevated multi-layer, GPS multipath causes p The projection is ambiguous, we introduce "floor preference" in S401 to prefer the road segment of the same elevation layer as the last valid segment to avoid pushing the car to the ground floor. The second type is caused by heavy rain, V_rt is significantly lower than the historical quantile, at this time w_r is naturally larger, V_b is lowered, Δt is not changed but the step is shortened, the extrapolated trajectory is more dense, which is convenient for subsequent fine alignment with the reverse backtracking. If the real-time road condition is not reported, w_r is degraded, V_b is more dependent and V_lim, the model extends with a conservative strategy, not blindly guessing congestion or smoothness.

[0094] The technical effects of the embodiment fall into three levels. On the data side, S401 integrates speed limits, road conditions, and individual speeds into a transparent speed generation logic, avoiding the dominance of a single source in prediction; on the geometric side, S402 uses curvature and turning cost to constrain displacement direction, preventing the predicted point from deviating from the drivable lane or making an excessive turn; on the process side, extrapolation to the midpoint and preservation of the evidence field provide an interface for the forward / backward fusion of S102, and subsequent weight calculation can reference the topological legality and speed source of each step in this segment, making it easier for engineers to understand why each point is placed there. Review the data base constructed by S401 / S402 and S201 / S202, all speed limits, ramp curvatures, and road speeds come from the same coordinate and time reference, the logic is closed loop, the engineers are at ease, and the next step of reverse backtracking is also easier to merge.

[0095] In an embodiment of the vehicle trajectory reconstruction and correction method based on spatio-temporal data fusion provided in the application, the following contents can be specifically included:

[0096] Step S501: From the floating car trajectory data, select the trajectory data passing through the same road segment at the same period as the target vehicle, extract the lane preference feature and speed variation law in the trajectory data, read the position coordinates and time stamp of the first discovered node after the break, determine the feasible reverse tracking path combining the one-way road restriction rules, construct a reverse trajectory prediction model based on the lane preference feature and speed variation law, and calculate the reverse tracking trajectory point;

[0097] Step S502: Pair the forward prediction trajectory point and the reverse tracking trajectory point according to the time stamp, calculate the spatial distance of each pair of trajectory points, set the fusion weight coefficient according to the spatial distance, and calculate the weighted average of the paired trajectory points to generate a fusion trajectory point. Connect the fusion trajectory points in chronological order to construct the trajectory break segment completion data.

[0098] Optionally, the embodiment revolves around S501 and S502, focusing on the evidence recovery and fusion of the broken section. We first select samples from the floating car database that pass through the same section at the same time as the target vehicle. To avoid mixing heterogeneous traffic, the selection criteria not only consider the overlap of road section ID and time, but also check the consistency of direction and the speed difference between adjacent points to exclude extreme sudden stops or abnormal jump points. After determining the samples, we extract two types of generalizable driving behavior features: lane preference and speed variation pattern. Lane preference is derived from the historical lane selection frequency and the yielding pattern at the weaving section, summarized as a probability vector such as "right lane straight probability" and "ramp entrance right turn tendency". The speed variation pattern is extracted from the deceleration-constant speed-acceleration pattern when entering the ramp, crossing the curve, and leaving the congestion section, and is normalized into a speed curve prototype that is independent of the time scale, retaining sensitivity to curvature and speed limit constraints. At this point, we read the coordinates and timestamp of the first discovery node N+ after the break, construct a reversible access upstream edge set based on the one-way rule, remove all edges that conflict with the direction or are forbidden to turn, and obtain the candidate path graph for reverse tracing.

[0099] The embodiment constructs a reverse trajectory prediction model on the candidate path graph, adopting a semi-parametric scheme rather than a black box. The state variables include position, direction, current lane type, and speed, and the input features come from the two types of behavior and topological speed limit mentioned above; the output is the displacement and lane selection at the next reverse time step. We use lane preference as a priori, scale match the speed curve prototype with the interval average speed of real-time road conditions to generate "expected speed-expected turning angle" guidance, and then use topological constraints as hard boundaries to obtain one or several reverse sequences R(t) with the highest score. If multiple branches with similar scores are encountered, the model will not arbitrarily choose one, but will retain the Top-K branches and record their cumulative confidence, and then let physical consistency decide in the subsequent fusion stage. To combat low-sample scenarios, we set a degenerate path: when there are not enough samples for the same section, the lane preference degenerates into a road grade weighting, and the speed curve degenerates into a linear combination of the speed limit and the average speed of the road conditions, still completing the backtracking within common sense constraints.

[0100] Before entering S502, we need to pair the forward prediction set F(t) and the reverse sequence R(t) on the same time axis. We construct a sampling grid of [t , t+] with a uniform step size Δt, and perform linear or spline interpolation on F and R respectively to ensure that there is a point pair at each sampling time. For each point pair, calculate the spatial distance d(t) and the direction deviation θ(t), which is used to avoid misjudgment of "close but opposite direction" at complex intersections. The weight setting depends on two metrics: geometric consistency and the credibility score of evidence sources (forward from topology and road conditions, reverse from behavior samples). We use adaptive distance decay and confidence stacking to assign weights, and limit the lower bound in extreme conflicts to prevent one side from completely losing voice.

[0101] This embodiment emphasizes the explainable linear synthesis on fusion calculation, without creating complex models. For each t, first calculate the geometric consistency part according to d(t) and θ(t), combine the forward topology confidence c_topo(t) and the reverse sample confidence c_beh(t) to obtain the temporary weight, and then normalize it, and then do weighted average on the two-point coordinates to generate the fusion point G(t). If Top-K reverse paths exist, we fuse each R_k(t) with F(t) respectively to obtain K candidate fusion sequences, and select the one with the lowest cost as the completion result of the broken section according to the overall acceleration smoothness and road adherence error criteria. This step logically gives an answer to "which branch is more like a road opened by a person", avoiding being dragged by local accidental noise.

[0102] This embodiment applies two corrections after fusion. The first one is the road adherence method, which projects G(t) to the drivable lane center and keeps the projection residual e_proj as a quality indicator. If the continuous residual rises, look back at the Top-K candidates in this interval and try to replace the suboptimal path to re-fuse; The second one is the dynamics consistency, which checks the speed and acceleration of adjacent time points. If there are unreasonable spikes, re-distribute the weights according to the principle of "who has weaker evidence gives way", and limit the upper bound of single-step displacement and the rate of change of turning angle to ensure that the trajectory can adhere to the road and not violate the common sense of vehicle motion. After correction, connect the fusion points in chronological order to form the trajectory broken section completion data, and archive the source weight, residual and attached road ID of each point for S103 interface display and manual review.

[0103] This embodiment considers two typical scenarios to illustrate the flexibility of the method. In the morning rush hour congestion under the ramp entrance, the forward prediction has a short step due to low road speed, while the reverse sample shows a strong tendency for the right lane to enter the ramp. During the fusion process, c_beh is high, and the trajectory deviates to the ramp earlier; The night traffic on the ring road main line is smooth, the sample is sparse, and the reverse degenerates to speed limit + road condition, c_topo dominates, and the trajectory goes straight along the main line to avoid the small branch without evidence. Both results conform to driving common sense and leave a clear explanation chain for the reviewer.

[0104] The technical effects of the embodiment mainly manifest in three aspects. The broken section is no longer connected by a straight line or pure geometric interpolation, but the "physically feasible forward" and the "reverse of group behavior" are connected on a unified time axis. If the conflict is large, the weight is reduced, and if the evidence is strong, the points are divided. The output is a curve with credentials. In places with multiple branches and insufficient evidence, multiple candidates are retained and selected by full cost to avoid being kidnapped by single-point noise. The fusion result is marked with a quality index and a source proportion. In the subsequent trajectory editing interface, engineers can quickly identify which points need to be checked, reducing invalid dragging. The data base of S501 / S502 and S201 / S202, the space-time unification of S301 / S302, and the forward generation of S102 are associated. The whole chain is like a tightened rope, and the force can be transmitted from the data source to the final curve point by point. The process is clear and can withstand review.

[0105] In an embodiment of the vehicle trajectory reconstruction and deviation correction method based on spatio-temporal data fusion provided in the application, the following contents can be specifically included:

[0106] Step S601: load the vector data of the geographic information road network, construct a road network rendering layer, draw topological elements such as road level, ramp position, and one-way identification on the road network rendering layer, read the node coordinate sequence in the trajectory broken section completion data, create a trajectory drawing layer above the road network rendering layer, convert the node coordinate sequence into Bezier curve control points, and generate a smooth trajectory line;

[0107] Step S602: create a trajectory editing canvas based on a grid layout, map the road network rendering layer and the trajectory drawing layer to the coordinate system of the editing canvas, mark the position coordinates of the trajectory nodes, construct a node dragging controller in the editing canvas, associate the node dragging controller with the trajectory node position, and establish a coordinate transformation response mechanism.

[0108] Optionally, the present embodiment is directed to steps S601 and S602, and the generated algorithm is completed to become a readable and editable visual interactive object. The engineering entrance loads the vector data of the city road network from the GIS server, and preferentially takes the hierarchical data: road center line (including direction), lane attribute, ramp element, one-way and no-turn rule. After loading, the "road network rendering layer" is constructed in the rendering pipeline, the road level is mapped into line width and color, the ramp entrance is marked with a special symbol, and the one-way is represented by arrow and transparent mask. In order to avoid lag caused by one-time drawing, the layer uses block tiles and multi-level cache, and only the tiles entering the view are refreshed when the view is zoomed in. Then the node coordinate sequence of the trajectory broken segment completion data output from S102 is read, a legality check (time monotonic, coordinate domain, road ID can be parsed) is first performed, and then the geographic coordinates are projected to the current view coordinate system. In order to make the line closer to the real driving trajectory of the vehicle, the node sequence is converted to Bezier curve control points: the direction vector and curvature of each node are calculated, the control handle length is determined, the straight line segment is shortened, and the curved section is appropriately lengthened to avoid excessive bending or "smoothing" of the corner. After the control points are generated, the trajectory drawing layer is superimposed on the road network rendering layer, and resolution adaptive sampling is adopted, the dense area is appropriately thinned, and the area with large curvature is densely sampled, so that the trajectory line is smooth and aesthetically consistent.

[0109] In order to let the user judge "where to check" at a glance, the present embodiment also superimposes node elements in the trajectory drawing layer, and the visual coding of the node comes from the confidence field and topological consistency score of the completion data. We mark high-confidence nodes as light-colored small dots and low-confidence nodes as dark-colored dots with borders, so that the user can pay attention to them first; at the ramp or complex intersection, the node will be attached with a miniature direction indicator, which is aligned with the road network arrow for visual inspection. When rendering, the bidirectional index of the node and the road segment is preserved: the node can find its road ID and the speed limit of the segment, and the map object can also find which nodes fall on it. The visual result after superimposition is a trajectory "sticking to the road" on the road network, which has both continuity and provides anchor points for editing.

[0110] The present embodiment enters S602 and starts to build an interactive editing canvas. We create a grid-based canvas in the interface container, and map the road network rendering layer and the trajectory drawing layer to the canvas coordinate system. The key to mapping is to maintain a reversible coordinate transformation stack: geographic coordinates→projected coordinates→screen coordinates, and scaling, translation and rotation actions are reflected on the transformation stack; for any mouse event, the screen coordinates can be inversely solved back to the geographic coordinates, avoiding interaction drift. When the canvas is initialized, the node graphical elements are placed according to the current view, their position coordinates are marked, and these elements are bound with corresponding geographic coordinates, road ID, confidence and other attributes to form a one-to-one mapping of "visual object-data object".

[0111] The soul of the editing is put in the "node drag controller". The controller registers pointer down, move, release events for each node. Record the original geographic coordinates and adjacent node index when down; read the screen displacement Δx, Δy provided by the event when moving, get the geographic displacement Δp by inverse transformation stack, calculate the new candidate coordinate p'. In order to not drag the trajectory away from the feasible road, we trigger the "coordinate transformation response mechanism" in real time during the movement, which contains three constraints: road fitting law (absorb p' to the nearest feasible lane center line, if the nearest line conflicts with the one-way direction, switch to the next feasible candidate), dynamic smoothing (evaluate the speed and acceleration of the front and rear sections, if there is a non-physical peak, automatically shrink Δp), curvature protection (limit the handle length in the curve area to prevent the corner from being straightened too much). Only do local recalculation when releasing: update the Bezier control points and sample points of the adjacent two sections of the node, avoid the lag caused by redrawing the whole curve, the interface feedbacks within a frame, and the interactive experience is natural.

[0112] This embodiment maintains a traceable editing log on the data side, records node ID, original coordinates, new coordinates, triggered constraint type and timestamp for each drag, and supports undo / redo. The log also retains the source weight (forward / backward fusion weight) of S102 stage, so that engineers can choose the modification range based on "evidence weight" when correcting. For complex intersections, single node dragging is often not enough, we provide segment-level tools: users select a segment of trajectory, translate or rotate by a small angle, the system allocates displacement and angle according to arc length, and the curvature distribution is approximately unchanged, which is suitable for handling overall offset or alignment after updating the base map. For areas with elevated / ground overlays, the canvas supports road layer filtering, and after selecting the target layer, the node adsorption searches only on the roads in that layer, avoiding the "sucked to the floor" embarrassment caused by Z layer confusion.

[0113] This embodiment takes into account the rendering capabilities of different terminals, and the canvas adopts a progressive detail strategy. When zooming in at the city level, the trajectory line is simplified to sparse skeleton points, and dragging is temporarily closed to avoid accidental touch; when zooming in at the block level, the nodes and control points are fully presented, and dragging is enabled. The transformation stack recalculates the sampling step after each view change to ensure that the line is smooth and does not consume too much GPU time. To facilitate verification, we pop up an information card when the node is hovered over, displaying the road ID, speed limit, one-way direction, fusion confidence, and the last editor's information, so users can determine whether to continue adjusting without switching panels.

[0114] In an embodiment of the vehicle trajectory reconstruction and correction method based on spatio-temporal data fusion provided in the present application, the following contents can be specifically included:

[0115] Step S701: Extract the node coordinate sequence from the trajectory broken section completion data, calculate the direction vector and distance between adjacent nodes, generate trajectory control points based on the third-order Bezier curve interpolation algorithm, connect the control points into a smooth trajectory line, mark the node position on the geographic information road network layer, draw the node icon, and establish a mapping relationship between the node icon and the coordinate data;

[0116] Step S702: Build an interactive layer, create a node dragging response area in the interactive layer, listen to the mouse press event, obtain the real-time coordinate change during dragging, update the node position according to the coordinate change, recalculate the trajectory control points, dynamically update the trajectory line shape, and record the adjusted node coordinate data.

[0117] Optionally, the embodiment is developed around steps S701 and S702, and focuses on converting the broken section completion data output by the algorithm into a beautiful trajectory line that can be fine-tuned by hand. We first read the node coordinate sequence {Pi} arranged in time sequence from the fusion result of S102, each Pi carrying geographic coordinates, road ID, and local confidence. To avoid direct connection with angles, we calculate the direction vector vi=normalize(Pi+1 Pi) and the distance di=||Pi+1 Pi|| of adjacent nodes in pairs. On this basis, control points are prepared for third-order Bezier interpolation. The method is to first estimate the normalized tangent ti=(vi 1+vi) of node i, and then give the control point offset hi=κ·min(di 1,di) according to the local scale, κ is the curve tightness parameter, which is derived from the road curvature and speed limit: κ is small in sharp curves and κ is large in straight roads, so as not to pull the line out of the road surface at the ramp. From this, we get the four-point group [Pi, Ci1, Ci2, Pi+1] of each section, where Ci1=Pi+ti·hi, Ci2=Pi+1 vi·hi. Connect all the four-point groups in order, and render the smooth trajectory line with adaptive step sampling resolution; and on the geographic information road network layer, drop a node icon for each Pi, the icon color gradually changes according to the confidence level, and the icon and the coordinate are bidirectionally mapped through a unique ID, which can be directly linked to the underlying data from the interface elements during subsequent interactive updates.

[0118] This embodiment does not separate visualization from data, and keeps a "geo-geometry" dual coordinate buffer while drawing the trajectory. Map zoom or rotation only affects the rendering coordinate, and the geo-coordinate is always kept as original, avoiding the accumulation of repeated projection errors. In order to make the curve more like "driven by the car", a small adjustment is made to each segment during sampling: the sampling point is projected to the center line of the attached road, and if the offset exceeds the threshold ε, it will converge to the center line along the normal, and at the same time record the offset residual e_proj of the point on the spline, which will be exposed as a risk prompt in the interface if the point needs to be adjusted again in S702.

[0119] This embodiment enters S702 to build interactive capabilities. We superimpose an interactive layer on top of the trajectory layer, creating a drag response area for each node icon, and the area size grows or shrinks adaptively with the current zoom level. The event pipeline listens to three types of actions: press, move, and release. When pressed, capture the node ID, its original geographic coordinates, and the adjacent segment index; when moved, read the change amount Δp_scr of the screen coordinates in real time, and use the current view matrix inverse to get the geographic increment Δp_geo, and get the candidate new coordinates P'i = Pi + Δp_geo. In order to not drag the point out of the common sense boundary, we do two constraint checks before updating: one is road feasibility, which attracts P'i to the nearest and direction-matched lane center line; the other is dynamic continuity, which calculates the speed and acceleration changes of the two adjacent segments after adjustment, and if it is out of bounds, it will throw P'i back to the feasible region boundary. The release event triggers local recalculation, which only generates new control points Ci1 / Ci2 for the affected two adjacent spline segments, and the trajectory line is locally refreshed, while the other parts remain unchanged, and the interaction delay is reduced.

[0120] This embodiment rigorously records the edited data. Each time the drag ends, we record the quadruple 〈node ID, original coordinates, new coordinates, constraint hit〉 and timestamp it, and enter the "edit log stack", supporting undo and redo. To reduce back and forth repetition, the system also writes the source weight (forward / backward ratio) of the node in S102 and the before / after value of e_proj in the log, so that engineers can judge whether this move has pulled the point from "model uncertainty" to "closer to the road" or just changed a curve that looks more pleasing to the eye.

[0121] This embodiment takes into account various field details, and the curve control point calculation is not a one-size-fits-all. For high-curvature ramps, we dynamically reduce kappa, and even fall back to piecewise circular arc approximation at extreme sharp turns to avoid the outer Bézier swing across adjacent lanes; under multi-level interchanges, the road ID of the node may be cut off the layer, and when dragging, the user once selects the target road level, the adsorption searches only in the center line of the layer to prevent adsorption to the ground road. Another common requirement is "segment-level fine-tuning": after the user selects a segment, the system will distribute the affine transformation to each node in the segment according to the arc length, and then refresh the control points according to the above rules to ensure the overall fidelity of the line shape and prevent the segment from bulging in the middle.

[0122] This embodiment gives a simple and traceable quality metric, which helps batch inspection outside the interface: Q = w_p·E_proj + w_c·C_curv. In the formula, Q is the quality index of the trajectory segment; E_proj is the mean or quantile of the projection offset of the sampling points of the segment, reflecting the "road adhesion degree"; C_curv is the penalty quantity of the curvature continuity of the segment, which can be approximated by the mean of the curvature difference of adjacent sampling points, reflecting "whether the line shape is too jagged"; w_p and w_c are weights, which are set according to road levels, and expressways pay more attention to E_proj, and urban branches pay more attention to C_curv. The parameter meanings are straightforward, and it is easy to land on engineering. The segment with high Q is preferentially put into the manual review list.

[0123] This embodiment does not pursue fancy effects, but makes it clear for both people and algorithms. S701 uses a spline that matches the road geometry to string the fusion nodes into an eye-pleasing line, and S702 uses controlled dragging to pull a small number of uncertain points back to the correct position. During the process, neither does the user have the freedom to draw lines at will, nor is the model output rigidly tied. Connect the two steps with the layer superposition of the previous S103 and the fusion trace of S501 / S502, every bend on the interface has a source, and every adjustment has a record. When reviewing in the future, you can trace the details of "why did you draw it like this at that time, and how did you change it later". This is more solid than simply outputting a trajectory.

[0124] In order to effectively solve the deficiencies of the traditional technology in data fusion, trajectory completion and visualization editing, and provide technical support for vehicle trajectory reconstruction, the present application provides an embodiment of a vehicle trajectory reconstruction and correction device based on spatio-temporal data fusion for implementing all or part of the contents of the vehicle trajectory reconstruction and correction method based on spatio-temporal data fusion, see Figure 2 , the vehicle trajectory reconstruction and correction device based on spatio-temporal data fusion specifically includes the following contents:

[0125] The data processing module 10 is used for collecting traffic monitoring data, dynamic road network topology data and traffic data, the traffic monitoring data includes a time stamp of a fisheye camera snapshot, target vehicle license plate information, position coordinates and a coordinate missing node marker, the dynamic road network topology data includes road grades, ramp structure parameters and single lane restriction rules, the traffic data includes floating car trajectory data and real-time traffic data, a reference control point is selected to establish a coordinate conversion matrix, the fisheye camera position coordinates are converted to geographic information road network coordinates, and the time stamps of the traffic monitoring data and the traffic data are calibrated based on a global positioning system standard time;

[0126] The trajectory completion module 20 is used for, for a trajectory broken section, taking the position coordinates, time stamps and speed information of the last known node before the break as a reference, combining road speed limits in the dynamic road network topology data and real-time traffic data to calculate forward prediction trajectory points, taking the first discovered node after the break as a starting point, combining driving features in the floating car trajectory data and single lane restriction rules to calculate reverse tracing trajectory points, and performing weighted average fusion on the forward prediction trajectory points and the reverse tracing trajectory points to generate trajectory broken section completion data.

[0127] The reconstruction and deviation correction module 30 is used for superimposing the trajectory broken section completion data on a geographic information road network layer, constructing a trajectory editing interface, reading node coordinates in the trajectory broken section completion data, drawing a trajectory line in the geographic information road network layer, marking the positions of the node coordinates, and supporting adjustment of the node coordinates through a drag operation.

[0128] As can be seen from the above description, the vehicle trajectory reconstruction and deviation correction device based on spatio-temporal data fusion provided in the embodiments of the present application can realize unified processing of data through coordinate conversion and time calibration by innovatively designing a multi-source data fusion model. A trajectory completion system is constructed, real-time traffic is combined with road network features to establish a reliable trajectory reconstruction mechanism. Visualization editing is introduced to ensure the accuracy of trajectory correction through interactive operation and coordinate adjustment. The method effectively solves the deficiencies of traditional technologies in data fusion, trajectory completion and visualization editing, and provides technical support for vehicle trajectory reconstruction.

[0129] From the hardware level, in order to effectively solve the deficiencies of traditional technologies in data fusion, trajectory completion and visualization editing, and provide technical support for vehicle trajectory reconstruction, the present application provides an embodiment of an electronic device for implementing all or part of the contents of the vehicle trajectory reconstruction and deviation correction method based on spatio-temporal data fusion, which specifically includes the following contents:

[0130] The system comprises a processor, memory, a communication interface, and a bus; wherein the processor, memory, and communication interface communicate with each other via the bus; the communication interface is used to realize information transmission between the vehicle trajectory reconstruction and correction device based on spatiotemporal data fusion and core business systems, user terminals, and related databases and other related devices; the logic controller can be a desktop computer, tablet computer, or mobile terminal, etc., and this embodiment is not limited to these. In this embodiment, the logic controller can be implemented with reference to the embodiments of the vehicle trajectory reconstruction and correction method based on spatiotemporal data fusion and the embodiments of the vehicle trajectory reconstruction and correction device based on spatiotemporal data fusion in the embodiments, the contents of which are incorporated herein, and repeated parts will not be described again.

[0131] It is understood that the user terminal may include smartphones, tablet computers, network set-top boxes, portable computers, desktop computers, personal digital assistants (PDAs), in-vehicle devices, smart wearable devices, etc. Among these, the smart wearable devices may include smart glasses, smartwatches, smart bracelets, etc.

[0132] In practical applications, parts of the vehicle trajectory reconstruction and correction method based on spatiotemporal data fusion can be executed on the electronic device side as described above, or all operations can be completed in the client device. The choice can be made based on the processing power of the client device and the limitations of the user's usage scenario. This application does not impose any limitations on this. If all operations are completed in the client device, the client device may further include a processor.

[0133] The aforementioned client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission with the server. The server may include a server on the task scheduling center side; in other implementation scenarios, it may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a distributed server structure.

[0134] Figure 3 This is a schematic block diagram illustrating the system configuration of the electronic device 9600 according to an embodiment of this application. Figure 3 As shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It is worth noting that... Figure 3 This is an example; other types of structures can also be used to supplement or replace this structure to achieve telecommunications functions or other functions.

[0135] In an embodiment, the vehicle trajectory reconstruction and deviation correction method based on spatio-temporal data fusion can be integrated into the central processor 9100. The central processor 9100 can be configured to control as follows:

[0136] Step S101: Collect traffic monitoring data, dynamic road network topology data, and traffic data. The traffic monitoring data includes a time stamp of a camera snapshot, target vehicle license plate information, position coordinates, and a coordinate missing node marker. The dynamic road network topology data includes road grades, ramp structure parameters, and single lane restriction rules. The traffic data includes floating car trajectory data and real-time traffic data. A reference control point is selected to establish a coordinate conversion matrix, and the camera position coordinates are converted to geographic information road network coordinates. The time stamps of the traffic monitoring data and the traffic data are calibrated based on the global positioning system standard time.

[0137] Step S102: For a trajectory broken section, the position coordinates, time stamps, and speed information of the last known node before the break are used as a reference, combined with the road speed limit in the dynamic road network topology data and the real-time traffic data to calculate forward prediction trajectory points. The first discovered node after the break is used as a starting point, combined with the driving characteristics in the floating car trajectory data and the single lane restriction rules to calculate reverse tracing trajectory points. The forward prediction trajectory points and the reverse tracing trajectory points are weighted and averaged to generate trajectory broken section completion data.

[0138] Step S103: The trajectory broken section completion data is superimposed on the geographic information road network layer to construct a trajectory editing interface. The node coordinates in the trajectory broken section completion data are read, and a trajectory line is drawn on the geographic information road network layer to mark the positions of the node coordinates. The node coordinates can be adjusted through a drag operation.

[0139] As can be seen from the above description, the electronic device provided by the embodiments of the present application innovatively designs a multi-source data fusion model, and realizes unified processing of data through coordinate conversion and time calibration. A trajectory completion system is constructed, and a reliable trajectory reconstruction mechanism is established in combination with road network characteristics and real-time traffic. Visualization editing is introduced, and the accuracy of trajectory correction is ensured through interactive operation and coordinate adjustment. This method effectively solves the deficiencies of traditional technologies in data fusion, trajectory completion, and visualization editing, and provides technical support for vehicle trajectory reconstruction.

[0140] In another embodiment, the vehicle trajectory reconstruction and deviation correction device based on spatio-temporal data fusion can be configured separately from the central processor 9100, for example, the vehicle trajectory reconstruction and deviation correction device based on spatio-temporal data fusion can be configured as a chip connected to the central processor 9100, and the vehicle trajectory reconstruction and deviation correction method based on spatio-temporal data fusion is realized through the control of the central processor.

[0141] like Figure 3 As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily need to include these components. Figure 3 All components shown; in addition, the electronic device 9600 may also include Figure 3 For components not shown, please refer to existing technologies.

[0142] like Figure 3 As shown, the central processing unit 9100, sometimes also referred to as a controller or operating control, may include a microprocessor or other processor device and / or logic device, which receives input and controls the operation of various components of the electronic device 9600.

[0143] The memory 9140 may be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It may store the aforementioned failure-related information, and also store a program for executing that information. The central processing unit 9100 may execute the program stored in the memory 9140 to perform information storage or processing, etc.

[0144] Input unit 9120 provides input to central processing unit 9100. Input unit 9120 may be, for example, a keypad or touch input device. Power supply 9170 provides power to electronic device 9600. Display 9160 displays images and text. Display may be, for example, an LCD display, but is not limited thereto.

[0145] The memory 9140 can be a solid-state memory, such as a read-only memory (ROM), random access memory (RAM), a SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROMs. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142 for storing application programs and function programs or processes for executing the operation of the electronic device 9600 via the central processing unit 9100.

[0146] The memory 9140 can further include a data storage 9143 for storing data such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. A driver storage 9144 of the memory 9140 can include various drivers of the electronic device for communication functions and / or for performing other functions of the electronic device (e.g., a messaging application, a phonebook application, etc.).

[0147] The communication module 9110 is a transmitter / receiver that transmits and receives signals via the antenna 9111. The communication module 9110 (transmitter / receiver) is coupled to the central processor 9100 to provide input signals and receive output signals, as in the case of a conventional mobile communication terminal.

[0148] Based on different communication technologies, a plurality of communication modules 9110 can be provided in the same electronic device, such as a cellular network module, a Bluetooth module, and / or a wireless local area network module, etc. The communication module 9110 (transmitter / receiver) is further coupled to the speaker 9131 and the microphone 9132 via the audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby implementing a conventional telecommunication function. The audio processor 9130 can include any suitable buffer, decoder, amplifier, etc. In addition, the audio processor 9130 is coupled to the central processor 9100, thereby enabling recording on the local device via the microphone 9132 and enabling playing of a sound stored on the local device via the speaker 9131.

[0149] The embodiment of the present application further provides a computer readable storage medium capable of implementing all steps of the vehicle trajectory reconstruction and deviation correction method based on space-time data fusion with the execution subject being a server or a client in the above-mentioned embodiments, and a computer program is stored on the computer readable storage medium. When the processor executes the computer program, all steps of the vehicle trajectory reconstruction and deviation correction method based on space-time data fusion with the execution subject being a server or a client in the above-mentioned embodiments are implemented. For example, when the processor executes the computer program, the following steps are implemented:

[0150] Step S101: Collect traffic monitoring data, dynamic road network topology data, and traffic data. The traffic monitoring data includes a camera snapshot timestamp, target vehicle license plate information, position coordinates, and a coordinate missing node marker. The dynamic road network topology data includes road grades, ramp structure parameters, and single-lane restriction rules. The traffic data includes floating car trajectory data and real-time traffic data. A reference control point is selected to establish a coordinate conversion matrix, the camera position coordinates are converted to geographic information road network coordinates, and the traffic monitoring data and traffic data timestamps are calibrated based on the global positioning system standard time.

[0151] Step S102: For the trajectory broken section, the last known node position coordinates, timestamp, speed information before the break are used as the reference, combined with the road speed limit in the dynamic road network topology data and the real-time traffic data to calculate the forward prediction trajectory point, and the first discovered node after the break is used as the starting point, combined with the driving characteristics in the probe vehicle trajectory data and the one-way restriction rules to calculate the reverse tracing trajectory point, and the forward prediction trajectory point and the reverse tracing trajectory point are weighted and averaged to generate trajectory broken section completion data;

[0152] Step S103: The trajectory broken section completion data is superimposed on the geographic information road network layer to construct a trajectory editing interface, the node coordinates in the trajectory broken section completion data are read, the trajectory line is drawn in the geographic information road network layer, the positions of the node coordinates are marked, and the node coordinates are adjusted through a drag operation.

[0153] From the above description, the computer readable storage medium provided by the embodiments of the application realizes unified processing of data through coordinate conversion and time calibration by innovatively designing a multi-source data fusion model. A trajectory completion system is constructed, real-time traffic is combined with road network characteristics to establish a reliable trajectory reconstruction mechanism. Visualization editing is introduced, and the accuracy of trajectory correction is ensured through interactive operation and coordinate adjustment. The method effectively solves the deficiencies of traditional technologies in data fusion, trajectory completion, and visualization editing, and provides technical support for vehicle trajectory reconstruction.

[0154] The embodiments of the application also provide a computer program product capable of realizing all steps of the vehicle trajectory reconstruction and deviation correction method based on spatio-temporal data fusion in the above-mentioned embodiments, which is executed by a processor to realize the steps of the vehicle trajectory reconstruction and deviation correction method based on spatio-temporal data fusion, for example, the computer program / instruction realizes the following steps:

[0155] Step S101: Collect traffic monitoring data, dynamic road network topology data, and traffic data, the traffic monitoring data includes a camera snapshot timestamp, target vehicle license plate information, position coordinates, and coordinate missing node markers, the dynamic road network topology data includes road grades, ramp structure parameters, and one-way restriction rules, the traffic data includes probe vehicle trajectory data and real-time traffic data, a reference control point is selected to establish a coordinate conversion matrix, the camera position coordinates are converted to geographic information road network coordinates, and the traffic monitoring data and traffic data timestamps are calibrated based on the global positioning system standard time;

[0156] Step S102: For the trajectory broken section, taking the position coordinates, time stamp and speed information of the last known node before the break as the reference, combining the road speed limit in the dynamic road network topology data and the real-time traffic data to calculate the forward prediction trajectory point, taking the first discovered node after the break as the starting point, combining the driving features in the floating car trajectory data and the one-way restriction rules to calculate the reverse tracing trajectory point, and performing weighted average fusion on the forward prediction trajectory point and the reverse tracing trajectory point to generate trajectory broken section completion data.

[0157] Step S103: Superimposing the trajectory broken section completion data on the geographic information road network layer to construct a trajectory editing interface, reading the node coordinates in the trajectory broken section completion data, drawing a trajectory line in the geographic information road network layer, marking the positions of the node coordinates, and supporting adjustment of the node coordinates through a drag operation.

[0158] As can be seen from the above description, the computer program product provided by the embodiments of the present application realizes unified processing of data through coordinate conversion and time calibration by innovatively designing a multi-source data fusion model. A trajectory completion system is constructed, a reliable trajectory reconstruction mechanism is established in combination with road network features and real-time traffic, and visual editing is introduced to ensure the accuracy of trajectory correction through interactive operation and coordinate adjustment. The method effectively solves the deficiencies of traditional technologies in data fusion, trajectory completion and visual editing, and provides technical support for vehicle trajectory reconstruction.

[0159] Those skilled in the art will understand that the embodiments of the present application can be provided as a method, device, or computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0160] The present application is described with reference to flowcharts and / or block diagrams of the method, device (apparatus), and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The device that implements the functions specified in one or more flows and / or blocks. Figure 1 The device that implements the functions specified in one or more flows and / or blocks.

[0161] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 one or more flow or blocks Figure 1 one or more flow or blocks

[0162] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions that are executed on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 one or more flow or blocks Figure 1 one or more flow or blocks

[0163] The principles and implementations of the present application are described in the specific embodiments, the above description of the embodiments is only used to help understand the method of the present application and its core idea; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation and application range will be changed, and the above description of the present application should not be understood as a limitation.

Claims

1. A vehicle trajectory reconstruction and deviation correction method based on spatiotemporal data fusion, characterized in that, The method comprises: Collecting traffic monitoring data, dynamic road network topology data and traffic data, the traffic monitoring data including camera snapshot timestamp, target vehicle license plate information, position coordinates and coordinate missing node marker, the dynamic road network topology data including road grade, ramp structure parameters and single lane restriction rules, the traffic data including floating car trajectory data and real-time traffic data, selecting a reference control point to establish a coordinate conversion matrix, converting the camera position coordinates to geographic information road network coordinates, and calibrating the timestamps of the traffic monitoring data and traffic data based on global positioning system standard time; For the trajectory broken section, taking the position coordinates, timestamps and speed information of the last known node before the break as the reference, combining the road speed limit in the dynamic road network topology data and the real-time traffic data to calculate the forward prediction trajectory point, including: extracting the position coordinates, timestamps and speed information of the last known node before the trajectory break, finding the road grid number corresponding to the position coordinates in the dynamic road network topology data, obtaining the road grade speed limit value of the road grid, reading the current average driving speed of the road grid from the real-time traffic data, and calculating the predicted reference speed by weighting the average driving speed and the speed limit value; based on the predicted reference speed, an uniform motion kinematics model is constructed, the position coordinates are substituted into the uniform motion kinematics model, the coordinate position at the next prediction time is calculated, the predicted coordinates are corrected according to the ramp curvature parameters of the road grid, the forward prediction trajectory point is generated, and the prediction calculation is repeated until the middle time of the broken section is reached; Taking the first discovered node after the break as the starting point, combining the driving characteristics in the floating car trajectory data and the single lane restriction rules to calculate the reverse tracing trajectory point, and performing weighted average fusion on the forward prediction trajectory point and the reverse tracing trajectory point to generate trajectory broken section completion data, including: selecting trajectory data of the same section through by the target vehicle in the same period from the floating car trajectory data, extracting lane preference characteristics and speed variation rules in the trajectory data, reading the position coordinates and timestamps of the first discovered node after the break, determining the feasible reverse tracing path combining the single lane restriction rules, constructing a reverse trajectory prediction model based on the lane preference characteristics and speed variation rules, and calculating the reverse tracing trajectory point; pairing the forward prediction trajectory point and the reverse tracing trajectory point according to the timestamps, calculating the spatial distance of each pair of trajectory points, setting a fusion weight coefficient according to the spatial distance, performing weighted average calculation on the paired trajectory points, generating a fusion trajectory point, and connecting the fusion trajectory points in time sequence to construct the trajectory broken section completion data; Superimposing the trajectory broken section completion data on the geographic information road network layer to construct a trajectory editing interface, reading the node coordinates in the trajectory broken section completion data, drawing a trajectory line in the geographic information road network layer, marking the position of the node coordinates, and supporting adjustment of the node coordinates through drag operation.

2. The trajectory reconstruction and rectification method based on spatiotemporal data fusion of a vehicle according to claim 1, characterized in that, The collection traffic monitoring data, dynamic road network topology data and traffic data, the traffic monitoring data contains the time stamp of the camera snapshot, the target vehicle license plate information, the position coordinates and the coordinate missing node mark, the dynamic road network topology data contains the road grade, the ramp structure parameter and the one-way restriction rule, the traffic data contains the floating car trajectory data and the real-time traffic data, including: Obtain the traffic monitoring data stream from the city traffic monitoring platform, analyze the device identification code, the snapshot timestamp, the license plate number, the latitude and longitude coordinate field in the data stream, mark the data node with missing latitude and longitude coordinates as invalid node, extract the road grade classification table, the ramp curvature parameter table and the one-way traffic rule table from the road network management system, construct the dynamic road network topology data structure, and establish the index mapping of the road network node and the connection relationship; Collect the latitude and longitude, speed and timestamp data uploaded by the vehicle positioning device from the floating car data platform, group and store the data according to the vehicle number, obtain the real-time traffic state data of the road grid from the map service platform, extract the average speed data in the road section, associate and save the average speed data with the road grid number, and establish the real-time traffic query index.

3. The trajectory reconstruction and rectification method based on spatiotemporal data fusion of a vehicle according to claim 1, characterized in that, The reference control point is selected to establish a coordinate conversion matrix, the camera position coordinates are converted to the geographic information road network coordinate system, and the timestamps of the traffic monitoring data and the traffic data are calibrated based on the global positioning system standard time, including: Obtain the control point coordinates of the landmark building from the city geographic information database, calculate the relative position relationship of the control point in the local coordinate system and the geographic information road network coordinate system, construct an affine transformation matrix, calculate the coordinate system conversion parameters according to the affine transformation matrix, apply the conversion parameters to the coordinate system conversion of the camera position coordinates, and generate the camera distribution data in the geographic information road network coordinate system; Read the standard time information of the global positioning system timing center, use the standard time information as the reference time, calculate the deviation value of the camera device timestamp in the traffic monitoring data and the vehicle-mounted device timestamp in the floating car trajectory data relative to the reference time respectively, modify the timestamp data according to the deviation value, and establish a unified time reference.

4. The trajectory reconstruction and rectification method based on spatiotemporal data fusion of claim 1, wherein, The trajectory broken segment completion data is superimposed to the geographic information road network layer to construct a trajectory editing interface, including: Load the vector data of the geographic information road network, construct a road network rendering layer, draw the road grade, ramp position, one-way identification and other topological elements on the road network rendering layer, read the node coordinate sequence in the trajectory broken segment completion data, create a trajectory drawing layer above the road network rendering layer, convert the node coordinate sequence to Bezier curve control points, and generate a smooth trajectory line; Create a trajectory editing canvas based on the grid layout, map the road network rendering layer and the trajectory drawing layer to the coordinate system of the editing canvas, mark the position coordinates of the trajectory nodes, construct a node dragging controller in the editing canvas, associate the node dragging controller with the trajectory node position, and establish a coordinate transformation response mechanism.

5. The trajectory reconstruction and rectification method based on spatiotemporal data fusion of claim 1, wherein, The node coordinates in the trajectory broken section completion data are read, a trajectory line is drawn in the geographic information road network layer, the positions of the node coordinates are marked, the node coordinates are adjusted through a drag operation, and the adjusting includes: Node coordinate sequences are extracted from trajectory broken section completion data, direction vectors and distances between adjacent nodes are calculated, trajectory line control points are generated based on a third-order Bezier curve interpolation algorithm, the control points are connected as a smooth trajectory line, node positions are marked in a geographic information road network layer, node icons are drawn, and a mapping relationship between the node icons and coordinate data is established; An interactive layer is constructed, a node drag response area is created in the interactive layer, a mouse press event is listened to, real-time coordinate changes in a drag process are obtained, node positions are updated according to the coordinate changes, trajectory line control points are recalculated, a trajectory line shape is dynamically updated, and adjusted node coordinate data are recorded.

6. A vehicle trajectory reconstruction and deviation correction device based on spatiotemporal data fusion, characterized in that, The device comprises: A data processing module is configured to collect traffic monitoring data, dynamic road network topology data and traffic data, the traffic monitoring data includes a camera snapshot timestamp, target vehicle license plate information, position coordinates and a coordinate missing node marker, the dynamic road network topology data includes a road grade, ramp structure parameters and a single lane restriction rule, the traffic data includes floating car trajectory data and real-time traffic data, a reference control point is selected to establish a coordinate conversion matrix, the camera position coordinates are converted to geographic information road network coordinates, and the timestamps of the traffic monitoring data and the traffic data are calibrated based on a global positioning system standard time. The trajectory completion module is used for the trajectory fracture section, taking the position coordinates, timestamp and speed information of the last known node before fracture as a reference, combining the road speed limit in the dynamic road network topology data and the real-time traffic data to calculate the forward prediction trajectory point, including: extracting the position coordinates, timestamp and speed information of the last known node before trajectory fracture, searching for the road grid number corresponding to the position coordinates in the dynamic road network topology data, obtaining the road grade speed limit value of the road grid, reading the current average driving speed of the road grid from the real-time traffic data, and calculating the prediction reference speed by weighting the average driving speed and the speed limit value; constructing a uniform speed kinematics model based on the prediction reference speed, substituting the position coordinates into the uniform speed kinematics model, calculating the coordinate position at the next prediction time, correcting the predicted coordinates according to the ramp curvature parameters of the road grid, generating the forward prediction trajectory point, and repeating the prediction calculation until the middle time of the fracture section is reached; taking the first discovered node after fracture as the starting point, combining the driving features in the probe vehicle trajectory data and the one-way restriction rules to calculate the backward tracing trajectory point, and performing weighted average fusion on the forward prediction trajectory point and the backward tracing trajectory point to generate trajectory fracture section completion data, including: screening the trajectory data of the same road section in the same period as the target vehicle from the probe vehicle trajectory data, extracting the lane preference features and speed change rules in the trajectory data, reading the position coordinates and timestamp of the first discovered node after fracture, determining the feasible backward tracing path combining the one-way restriction rules, constructing a backward trajectory prediction model based on the lane preference features and speed change rules, and calculating the backward tracing trajectory point; pairing the forward prediction trajectory point and the backward tracing trajectory point according to the timestamp, calculating the spatial distance of each pair of trajectory points, setting the fusion weight coefficient according to the spatial distance, performing weighted average calculation on the paired trajectory points, generating the fusion trajectory point, and connecting the fusion trajectory points in time sequence to construct the trajectory fracture section completion data. The reconstruction and deviation correction module is used for superimposing the trajectory fracture section completion data to the geographic information road network layer, constructing a trajectory editing interface, reading the node coordinates in the trajectory fracture section completion data, drawing a trajectory line in the geographic information road network layer, marking the position of the node coordinates, and supporting adjustment of the node coordinates through a drag operation.

7. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the steps of the vehicle trajectory reconstruction and deviation correction method based on spatio-temporal data fusion of any one of claims 1 to 5.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the vehicle trajectory reconstruction and deviation correction method based on spatio-temporal data fusion of any one of claims 1 to 5.

Images