A smart wharf multi-information source vehicle positioning data fusion and correction method

Abstract

The application discloses a kind of wisdom wharf multi-information source vehicle positioning data fusion and correction method, belong to wisdom wharf technical field.The method takes truck operation cycle as processing unit, collects four kinds of information source data of electronic fence, RTG camera, mobile phone GPS and TOS system;Correction layer uses the double-layer strategy of separation of coarse adjustment and fine adjustment, coarse adjustment is removed by weighted voting to remove data point, and fine adjustment is covered by business semantics using TOS operation quality score;Fusion layer constructs the credible weight matrix of field×information source, takes the maximum weight information source value to discrete label field, and carries out weighted average to continuous numerical field;Inference layer, when positioning signal is missing, fuses TOS target berth, motion state and stack area constraint, uses end-point constraint deceleration approach model to complete trajectory in limited time, the present application solves the problems of large difference in multi-source data characteristics, operation context fragmentation, field-level fusion missing and large trajectory deviation in signal blind area.

Description

Technical Field

[0001] This invention relates to the field of intelligent terminal technology, specifically a method for fusing and correcting vehicle positioning data from multiple information sources in an intelligent terminal. Background Technology

[0002] In smart terminal operations, real-time location information of container trucks is the fundamental data support for dispatching systems, operating equipment (such as RTGs and gantry cranes), and safety management systems. Accurate and continuous vehicle positioning trajectories are of great significance for improving yard operation efficiency and reducing the risk of safety accidents.

[0003] Currently, terminals typically use multiple information sources to obtain truck location data, including but not limited to: electronic fence systems based on RFID or geomagnetic sensors (providing area-level entry and exit events and work number binding), RTG camera recognition systems based on machine vision (providing license plate numbers, container numbers, and relative position coordinates), GPS / BeiDou positioning modules on truck drivers' mobile terminals (providing continuous position coordinates and vehicle speed), and the terminal operating system TOS (providing operational instructions, target bay locations, and other business semantic information).

[0004] However, the above-mentioned multiple information sources have the following technical problems in practical applications: First, the location data from various information sources differ significantly in terms of accuracy, continuity, update frequency, and reliability. Electronic fences can only provide regional event triggering information and cannot output continuous trajectories; RTG cameras are affected by field-of-view obstruction, lighting changes, and algorithm recognition confidence levels, resulting in intermittent location output with inherent uncertainties; mobile phone GPS signals are easily lost in container stacking areas of the yard due to the Faraday cage effect, and there is a location blind spot in the middle area of ​​the stacking zone (approximately 50 meters); the TOS system provides business-level semantic information and does not output real-time location itself. These differences in data characteristics among the aforementioned information sources make data fusion difficult, and direct splicing or simple weighted averaging can lead to problems such as trajectory jumps and logical contradictions.

[0005] Second, existing fusion methods often use fixed time slices as processing units, disrupting the continuity of the operational context. Traditional fusion algorithms such as Kalman filtering and particle filtering process each frame of data independently at fixed time steps (e.g., per second), ignoring the business semantic constraints within the truck operation cycle (from receiving the TOS instruction to completing loading and unloading). Positioning data from different times within the same operation should ideally have consistent target location, driving path, and destination parking requirements. However, fixed-time-slice processing cannot share this operational-level prior knowledge, leading to a disconnect between the fusion result and the business semantics (e.g., the fused location does not match the target location assigned by the TOS).

[0006] Third, the reliability of different information sources varies in discrete label fields (such as truck operation number, license plate number, and container number) and continuous numerical fields (such as location coordinates and vehicle speed). However, existing methods typically assign a single overall weight to the same information source, failing to achieve field-level differentiated processing. For example, mobile phone GPS has high accuracy in vehicle speed measurement, but is susceptible to multipath effects in location coordinates; RTG cameras have high resolution in location recognition, but may misjudge license plate numbers due to image angle. The lack of a field-level weight allocation mechanism causes the fusion result to deviate from the optimal information source in some fields.

[0007] Fourth, when the positioning signal is completely missing in the middle area of ​​the stack area (e.g., the mobile phone GPS signal is lost and the RTG camera has no field of view), existing methods mostly use linear interpolation or uniform extrapolation to complete the trajectory. These purely mathematical methods do not integrate the TOS business objective (target position) and the physical constraints of the stack area (speed limit, channel width), resulting in a large deviation between the completed trajectory and the actual driving path, and failing to meet the operational requirements of accurate parking at the destination.

[0008] Fifth, existing positioning correction methods typically employ a single knowledge domain processing strategy (such as median filtering based on data noise statistics or majority voting based on temporal consistency), failing to separate the removal of data layer anomalies from the consistency verification of the business semantic layer. When electronic fences or RTGs generate drifting data points due to environmental interference, simple filtering or voting cannot distinguish between truth value drift and changes in the operation path, leading to unreliable correction results.

[0009] In summary, existing methods for fusing and correcting vehicle positioning data from multiple information sources at smart terminals have shortcomings in terms of operational context continuity, field-level differentiated fusion, business constraint-driven trajectory extrapolation, and cross-knowledge domain two-layer correction. Summary of the Invention

[0010] The technical problem to be solved by the present invention is to provide a positioning data fusion and correction method that can use the work cycle as the processing unit, integrate multi-source heterogeneous data and perform field-level weighting, and combine TOS business semantics to complete the trajectory, which can solve the problems in the prior art.

[0011] This invention is achieved through the following technical solution: A method for vehicle positioning data fusion and correction from multiple information sources in a smart terminal, comprising a data acquisition layer, which includes four types of information sources: electronic fence (S1), RTG camera (S2), mobile phone GPS (S3), and TOS system (S4). The correction layer includes coarse adjustment and fine adjustment. The coarse adjustment layer performs weighted voting on the container location information reported by the electronic fence (S1), RTG camera (S2), and mobile phone GPS (S3) in the time-series dimension and outputs the coarse adjustment result. The fine adjustment layer introduces the operation context information from the TOS system (S4), performs business semantic level coverage correction through operation quality score Q(t), and outputs the fine adjustment result to the fusion layer.

[0012] A further technical solution, the working method of the correction layer is as follows: First, receive real-time positioning data from four types of information sources; second, calculate the reliability score of each information source; third, perform weighted voting within a sliding window; fourth, obtain the coarse adjustment result; fifth, judge the TOS operation quality score. If the score is greater than the operation quality confidence threshold, the TOS system will use the estimate that the truck should cover the position of the coarse adjustment result based on the current operation instruction. If the score is less than the operation quality confidence threshold, the coarse adjustment result will be maintained; sixth, output the fine adjustment result based on the judgment in the fifth step; seventh, send the fine adjustment result from the sixth step to the fusion layer.

[0013] Further technical solutions include a fusion layer comprising discrete field fusion, continuous field fusion, and timestamp alignment. Discrete field fusion processes four discrete label fields: truck operation number, license plate number, container number, and operation status. The fusion result is the output value of the source with the highest confidence weight among all valid information sources. Continuous field fusion processes three continuous numerical fields: location coordinates, driving direction, and vehicle speed. The fusion result is a weighted average of the output values ​​of each valid information source. Timestamp alignment uses the system's unified clock as a reference to align the data from each information source to the same time coordinate system, thus obtaining the positioning data output by the fusion layer.

[0014] A further technical solution involves using a target-oriented endpoint-constrained deceleration approach motion model in the simulation layer to perform finite-time trajectory simulation. When the system detects that the positioning data output by the fusion layer is continuously missing and there are no obstructing road sections or closed areas in front of the operation location in the stack area, the simulation layer is triggered to perform short-range simulation within a finite time window, so that the simulation results are smoothly connected with the known positioning data before and after the missing segment, and finally the output layer outputs a continuous positioning trajectory.

[0015] A further technical solution, the working method of the deduction layer is as follows: First, it detects that there are continuous gaps in the positioning data output by the fusion layer and that there are no obstructing road sections or closed areas in front of the operation location in the stack area; Second, it obtains the last known position p(t0), speed v(t0), and driving direction when the missing area is located; Third, it obtains the target position of the truck currently operating from the TOS system. Fourth step, initialize the simulation model; fifth step, calculate the guiding acceleration. If it is positive, proceed to step six; if it is negative, do not output the derivation trajectory. Step six: Perform one derivation step. Step 7: Triple constraint check. If it passes directly, proceed to step 8; if the speed exceeds the limit, limit the vehicle speed to [specific limit]. Then proceed to step eight; if the passage crosses the boundary, correct the vehicle's lateral position and proceed to step eight; in step eight, determine whether the target position has been reached or the simulation time has exceeded the limit. If so, proceed to step nine; otherwise, return to step five; in step nine, output the simulation trajectory, smoothly connect it with the existing positioning data, and output the complete positioning result.

[0016] A further technical solution is proposed: assuming N information sources participate in voting at time t (the TOS system does not participate in the coarse adjustment layer), the i-th information source report set card is located at position β. Its dynamic reliability score is Weighted voting functions include: in, This represents the weighted voting result of the reported bit b at time t. [·] is an indicator function that takes a value of 1 when the condition is true and 0 otherwise. Scoring the reliability of each information source; Within the sliding time window [t-ΔT, t], the beta with the highest cumulative weighted votes for beta b is taken as the coarse adjustment result. : For the reporting position b in Time-weighted voting results; Reliability scores of each information source The following combination method is used to determine: Electronic fence (S1): Under normal conditions, a fixed empirical value r1 = r0 is used. When an abnormal fence event trigger interval is detected, the reliability is downgraded. RTG camera (S2): Based on the confidence score output by the recognition algorithm, , where confidence is the confidence score output by the algorithm; w is a hyperparameter between [0, 1]; Mobile GPS (S3): Based on the satellite signal quality index HDOP value r3(t) = f(HDOP(t)), where f is a monotonically decreasing mapping function; TOS (S4): Does not participate in the coarse adjustment layer weighted voting process, but only intervenes in the fine adjustment layer.

[0017] Further technical solutions and rules for fine-tuning the layer are as follows: when At that time, fine-tuning results, ,otherwise: ; in Assign a confidence threshold for job quality. The TOS system estimates the position of the truck based on the current work instructions.

[0018] A further technical solution involves constructing a trusted weight matrix at the fusion layer: W Where F=7 represents the number of fields to be merged, and S=4 represents the number of information sources; Each element in the matrix This represents the credibility weight value of the s-th information source in the f-th field.

[0019] Further technical solutions employ two different fusion strategies in the fusion layer, depending on the data type of the fields: Discrete field fusion strategy: For the four discrete label fields of truck operation number, license plate number, container number, and operation status, the fusion result is the output value of the source with the highest credibility weight among all valid sources. The physical meaning of this strategy is: in determining discrete labels, trust the most reliable source of information.

[0020] Continuous field fusion strategy: For the three continuous numerical fields of location coordinates, driving direction, and vehicle speed, the fusion result is a weighted average of the output values ​​of each valid information source. Only sources with valid outputs are included in the summation; invalid sources are not included in the fusion calculation of this field.

[0021] Timestamp alignment: Based on the unified system clock, the data from each information source are aligned to the same time coordinate system. The timestamp itself does not participate in the weight fusion.

[0022] Further technical solutions and the input information required for the deduction include: Initial state: the last known position p(t0), speed v(t0), and direction of travel of the truck when it enters the missing area; Target Status: The target position of the current operation of this truck in the TOS system. ; Constraints: Speed ​​limit in the storage area is 30 km / h (≈8.33 m / s), storage area access width is limited, and parking requirements at the end point are required; The derivation model uses the following equations of motion: Where δ is the time step of the deduction; To guide acceleration, the target position is... With the current simulation position The relative relationship was calculated.

[0023] Further technical solutions to guide acceleration The calculation is subject to three constraints: Endpoint constraint: The truck must ultimately stop precisely below the target position, i.e., when p(t) → When v(t) → 0; Speed ​​constraint: The speed at any point during the simulation must not exceed the reactor zone speed limit. ; Channel constraints: The stacking area channel is a straight channel, and the lateral offset of the truck is limited by the channel width.

[0024] The beneficial effects of this invention are as follows: First, it proposes a multi-source positioning data fusion architecture based on the truck operation cycle as the processing unit. This architecture uses a complete truck operation (from receiving the TOS operation instruction to completing the loading / unloading operation) as the basic unit of data processing, replacing the traditional architecture that uses fixed time slices as processing units. The three processing stages—correction, fusion, and deduction—are all completed within the same operation context, enabling the entire data processing chain to share operation-level prior knowledge and business constraints. This architectural innovation forms the basis for the subsequent three algorithmic innovations and holds a leading position in the overall design.

[0025] II. A two-layer adjustment and correction method separating the data layer and the business semantic layer is proposed, employing a cross-knowledge domain two-layer correction strategy that separates coarse adjustment and fine adjustment. In the data knowledge domain, the coarse adjustment layer uses a weighted voting mechanism with dynamic reliability scoring of the information source to vote on the temporal consistency of discrete bit labels, eliminating drifting data points. In the business semantic knowledge domain, the fine adjustment layer uses the TOS job quality score as a criterion for coverage correction. The two layers correspond to different knowledge domains and are non-mergeable, which is a core technical feature distinguishing them from single-knowledge-domain two-step structures such as Kalman filtering.

[0026] Third, a field-level differentiated trusted weight fusion method based on field data type classification is proposed. This method constructs a two-dimensional trusted weight matrix (7 fields × 4 sources) and employs a differentiated fusion strategy based on field data type: discrete label fields take the source with the highest weight, while continuous numerical fields use a weighted average. The same information source has different weights on different fields, achieving a dual refinement of "field type classification + field-level weight," which is the core technical feature distinguishing it from all existing source-level overall weighting schemes.

[0027] IV. A Finite-Time Trajectory Completion Method Integrating TOS Business Constraints: This paper proposes a finite-time extrapolation method based on a target-oriented endpoint constraint deceleration approach motion model. Addressing a positioning blind zone of approximately 50 meters (about 8 B) in the middle of the stack area, it combines TOS target B position information, initial motion state, and stack area physical constraints (speed limit 30 km / h, channel width) to perform short-range trajectory extrapolation within a finite time window (approximately 6 seconds). This method integrates three information sources: business objectives, motion equations, and scene constraints, which is a core technical feature distinguishing it from purely mathematical methods such as linear interpolation and uniform extrapolation.

[0028] Fifth, the advantage of using the truck operation cycle as the processing unit is that all time steps within the same operation share operation-level prior knowledge, which makes the correction, fusion and deduction processes have consistent business semantic constraints, avoiding the problem of business context fragmentation between different time steps when using fixed time slices as units. Attached Figure Description

[0029] For ease of explanation, the present invention will be described in detail below with reference to specific embodiments and accompanying drawings.

[0030] Figure 1 This is a schematic diagram of the overall system architecture; Figure 2 A flowchart illustrating a two-layer correction strategy that separates coarse and fine adjustments for the correction layer; Figure 3 This is a schematic diagram of the complete processing flow of the deduction layer. Detailed Implementation

[0031] like Figures 1-2 As shown, the present invention will be described in detail. For ease of description, the directions mentioned below are defined as follows: the directions of up, down, left, right, front, and back mentioned below are the same as... Figure 1 The directions of the projection relationship are consistent in all directions: up, down, left, right, front, and back.

[0032] Example 1: A method for vehicle positioning data fusion and correction from multiple information sources at an intelligent terminal according to the present invention includes a data acquisition layer, a correction layer, a fusion layer, a deduction layer, and an output layer. Taking the truck operation cycle as the basic processing unit, it accesses positioning data from four types of information sources and adopts a three-stage serial processing flow of "correction, fusion, and deduction" to output accurate, continuous, and semantically consistent truck positioning results.

[0033] Beneficially, the four information sources are: electronic fence (S1), which detects trucks entering and leaving the fenced area through RFID readers or geomagnetic sensors, providing regional positioning and identification of truck operation numbers based on RFID binding; RTG camera (S2), which uses computer vision algorithms to identify truck license plate numbers, container numbers and relative positions through industrial cameras installed on gantry cranes, and outputs an additional identification confidence score; mobile phone GPS (S3), which uses the GPS / BeiDou positioning module of the truck driver's mobile terminal to provide vehicle position coordinates and speed information, the reliability of which is affected by satellite signal quality (HDOP value); and TOS system (S4), which records the truck's operation instruction information, including target location, operation type, operation sequence and other business semantic data.

[0034] Beneficially, there are a total of 7 fields for fusion, which are divided into three categories according to data type: discrete label fields include truck operation number, license plate number, and container number; continuous numerical fields include location coordinates, driving direction, and vehicle speed; and there is also a timestamp field, which is aligned with the system's unified clock and does not participate in weight fusion.

[0035] Example 2, based on Example 1, further specifies the following: The correction layer adopts a two-layer correction strategy that separates coarse adjustment and fine adjustment. The coarse adjustment uses a weighted voting mechanism, while the fine adjustment uses business semantic coverage.

[0036] Beneficially, the coarse adjustment layer performs weighted voting on the location information reported by the electronic fence (S1), RTG camera (S2) and mobile phone GPS (S3) in the time dimension, and identifies and removes drift data points caused by Faraday cage effect or algorithm error.

[0037] Beneficially, assuming there are N information sources participating in the voting at time t (the TOS system does not participate in the coarse adjustment layer), the i-th information source report set card is located at position β. Its dynamic reliability score is Define a weighted voting function: in, This represents the weighted voting result of the reported bit b at time t. [·] is an indicator function that takes a value of 1 when the condition is true and 0 otherwise. Scoring the reliability of each information source; Within the sliding time window [t-ΔT, t], the beta with the highest cumulative weighted votes for beta b is taken as the coarse adjustment result. : For the reporting position b in Time-weighted voting results; Reliability scores of each information source The following combination method is used to determine: Electronic fence (S1): Under normal conditions, a fixed empirical value r1 = r0 is used. When an abnormal fence event trigger interval is detected (significantly deviating from the normal frequency), the reliability is downgraded. RTG camera (S2): Based on the confidence score output by the recognition algorithm, , where confidence is the confidence score output by the algorithm; w is a hyperparameter between [0, 1], set by human experience to smooth out the problem of the algorithm being overconfident; Mobile GPS (S3): Based on the satellite signal quality index HDOP value r3(t) = f(HDOP(t)), where f is a monotonically decreasing mapping function. The larger the HDOP value, the worse the geometric precision factor and the lower the reliability. TOS (S4): Does not participate in the coarse adjustment layer weighted voting process, but only intervenes in the fine adjustment layer.

[0038] The weighted voting mechanism uses discrete bit labels with spatial semantics as the voting object, rather than continuous numerical values. The voting process incorporates dynamic reliability scores for each information source, making it a weighted voting method rather than a simple majority vote. This method is fundamentally different from median filtering for continuous numerical values ​​and majority voting decoding in the communications field for semantically unsymmetric data.

[0039] Beneficially, the fine-tuning layer incorporates job context information from the TOS system and performs business semantic-level coverage correction through job quality score Q(t). The TOS job quality score Q(t) is defined, which comprehensively considers the completeness of the current job instruction, the clarity of the target bit, and the rationality of the job status. The fine-tuning rules are as follows: when At that time, fine-tuning results, ,otherwise: ; in Assign a confidence threshold for job quality. The TOS system estimates the position of the truck based on the current work instructions.

[0040] The fine-tuning layer addresses the issue of business semantic consistency—ensuring that the corrected location data is semantically consistent with the TOS job instructions. This belongs to a different knowledge domain than the data noise problem addressed by the coarse-tuning layer, and the two layers are not merging.

[0041] Example 3, based on any one of Examples 1 to 2, further defines the following: The fusion layer includes discrete field fusion, continuous field fusion, and timestamp alignment. Discrete field fusion uses the maximum weight source value, while continuous field fusion uses a weighted average.

[0042] The fusion layer constructs a reliable weight matrix: W Where F=7 represents the number of fields to be merged, and S=4 represents the number of information sources; Each element in the matrix This represents the credibility weight value of the s-th information source in the f-th field.

[0043] Beneficially, two different fusion strategies are adopted based on the different data types of the fields: Discrete field fusion strategy: For the four discrete label fields of truck operation number, license plate number, container number, and operation status, the fusion result is the output value of the source with the highest credibility weight among all valid sources. The physical meaning of this strategy is: in determining discrete labels, trust the most reliable information source and avoid making meaningless numerical averages of discrete values.

[0044] Continuous field fusion strategy: For the three continuous numerical fields of location coordinates, driving direction, and vehicle speed, the fusion result is a weighted average of the output values ​​of each valid information source. Only sources with valid outputs are included in the summation; invalid sources (such as GPS missing in the middle of the stack area) are not included in the fusion calculation of this field.

[0045] Timestamp alignment: Based on the unified system clock, the data from each information source are aligned to the same time coordinate system. The timestamp itself does not participate in the weight fusion.

[0046] The core principle of field-level differentiated weight allocation is to assign differentiated weights based on the data acquisition mechanism and precision characteristics of each information source on a specific field. The specific weight allocation is shown in the table below: The “—” indicates that the information source does not provide data for this field and does not participate in the fusion; the electronic fence (S1) has a high weight in the operation number because RFID is bound to the operation number mapping relationship; the RTG camera (S2) has a high weight in the location because it has high spatial resolution and visual recognition capability; the mobile phone GPS (S3) has a high weight in the vehicle speed because GPS Doppler speed measurement accuracy is good; the TOS system (S4) has accurate business information in the truck operation number and container number, so its credibility is high.

[0047] Example 4, based on any one of Examples 1 to 3, further defines the following: the simulation layer is based on the finite time trajectory completion of the TOS business constraints. In response to the problem that the positioning signal may be completely missing in the last 50 meters (about 8 B, 1 B ≈ 6 m) when the truck enters the middle area of ​​the stacking area, the simulation layer adopts the target-oriented endpoint constraint deceleration approach motion model to perform finite time trajectory simulation.

[0048] When the system detects that the positioning data output by the fusion layer is continuously missing (i.e., all information sources have no valid location output within a certain period of time) and there are no obstructing road sections or restricted areas in front of the operation location in the stack area, the inference layer is triggered.

[0049] Beneficially, the input information required for the deduction includes: Initial state: the last known position p(t0), speed v(t0), and direction of travel of the truck when it enters the missing area; Target Status: The target position of the current operation of this truck in the TOS system. ; Constraints: Speed ​​limit in the storage area is 30 km / h (≈8.33 m / s), width limit of the storage area passage is limited, and parking requirements at the end point are required.

[0050] Beneficially, the deductive model employs the following equations of motion: Where δ is the time step of the deduction; To guide acceleration, the target position is... With the current simulation position The relative relationship was calculated.

[0051] Benefically, the calculation of guided acceleration is subject to three constraints simultaneously: Endpoint constraint: The truck must ultimately stop precisely below the target position, i.e., when p(t) → When v(t) → 0; Speed ​​constraint: The speed at any point during the simulation must not exceed the reactor zone speed limit. = 8.33 m / s; Channel constraints: The stacking area channel is a straight channel, and the lateral offset of the truck is limited by the channel width.

[0052] The time frame for the simulation is strictly limited to a finite time window. Based on a missing segment of 50 meters and an average vehicle speed of approximately 8 m / s, the simulation time is estimated to be about 6 seconds.

[0053] Conducting short-range simulations within a limited time window ensures that the deviation between the simulated trajectory and the actual trajectory is within an acceptable range.

[0054] The output layer smoothly connects the inference results with the known positioning data before and after the missing segment, and finally outputs a continuous positioning trajectory.

[0055] Example 5, based on any one of Examples 1 to 4, further defines the following: The overall data processing flow of the present invention takes a complete truck operation (from receiving the TOS operation instruction to completing the loading / unloading operation) as the basic unit, and the above-mentioned three processing stages of correction, fusion, and deduction are all completed serially in the same operation context.

[0056] When the TOS system assigns a new job instruction to a truck, the system creates a job-level data processing context, loading prior information such as the target location, job type, and job sequence. Subsequently, throughout the entire execution cycle of the job, all positioning data collected from the four types of information sources is corrected and integrated within this job context. The fine-tuning layer and the deduction layer can directly reference the TOS information in the job context for business semantic constraints and target-oriented deduction. Once the job is completed, the system closes the current processing context and awaits the arrival of the next job instruction.

[0057] The advantage of using the truck operation cycle as the processing unit is that all time steps within the same operation share operation-level prior knowledge, which makes the correction, fusion and deduction processes have consistent business semantic constraints, avoiding the problem of business context fragmentation between different time steps when using fixed time slices as units.

[0058] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions conceived without creative effort should be included within the scope of protection of the present invention; therefore, the scope of protection of the present invention should be determined by the scope defined in the claims.

Claims

1. A method for fusing and correcting vehicle positioning data from multiple information sources at a smart terminal, comprising a data acquisition layer, the data acquisition layer including four types of information sources: electronic fence (S1), RTG camera (S2), mobile phone GPS (S3), and TOS system (S4), characterized in that, The correction layer includes coarse adjustment and fine adjustment. The coarse adjustment layer performs weighted voting on the location information of the electronic fence (S1), RTG camera (S2) and mobile phone GPS (S3) in the time dimension and outputs the coarse adjustment result. The fine adjustment layer introduces the job context information in the TOS system (S4), performs business semantic level coverage correction through job quality score Q(t), and outputs the fine adjustment result to the fusion layer. 2.The method of claim 1, wherein: The correction layer operates as follows: First, it receives real-time positioning data from four types of information sources; second, it calculates the reliability score of each information source; third, it performs weighted voting within a sliding window; fourth, it obtains the coarse adjustment result; fifth, it judges the TOS operation quality score. If the score is greater than the operation quality confidence threshold, the TOS system uses the estimate that the truck should cover the position in the coarse adjustment result based on the current operation instruction; if the score is less than the operation quality confidence threshold, the coarse adjustment result is maintained; sixth, it outputs the fine adjustment result based on the judgment in step five; seventh, it sends the fine adjustment result from step six to the fusion layer.

3. The method according to claim 1 or 2, characterized in that: At time t, N information sources participate in voting, the ith information source reports that the set card is located in the bay , and the dynamic reliability score is , and the weighted voting voting function comprises: wherein, is the weighted voting result of the reporting bin b at time t, [·] is an indicator function, which takes the value 1 when the condition is true, and 0 otherwise, is the dynamic reliability score; In the sliding time window [t-ΔT, t], the b position with the highest cumulative weighted vote number is taken as the rough adjustment result : For the reporting position b in Time-weighted voting results; Source reliability score Is determined using the following mix: Electronic fence (S1): Under normal conditions, a fixed empirical value r1 = r0 is used. When an abnormal fence event trigger interval is detected, the reliability is downgraded. RTG camera (S2): based on the confidence score output by the recognition algorithm, where confidence is the confidence score output by the algorithm; w is a hyperparameter between [0, 1]; Mobile GPS (S3): Based on the satellite signal quality index HDOP value r3(t) = f(HDOP(t)), where f is a monotonically decreasing mapping function; TOS (S4): Does not participate in the coarse adjustment layer weighted voting process, but only intervenes in the fine adjustment layer.

4. The method according to claim 1 or 2, characterized in that: The rules for fine-tuning the layers are as follows: when At that time, fine-tuning results, ,otherwise: ; wherein is a job quality confidence threshold, is a TOS system estimate of where the tractor should be in the bay based on current job instructions.

5. The method according to claim 1 or 2, characterized in that: The fusion layer includes discrete field fusion, continuous field fusion, and timestamp alignment. Discrete field fusion processes four discrete label fields: truck operation number, license plate number, container number, and operation status. The fusion result is the output value of the source with the highest confidence weight among all valid sources. Continuous field fusion processes three continuous numerical fields: location coordinates, driving direction, and vehicle speed. The fusion result is the weighted average of the output values ​​of each valid source. Timestamp alignment uses the system's unified clock as a reference to align the data from each source to the same time coordinate system, thus obtaining the positioning data output by the fusion layer.

6. The method for fusing and correcting vehicle positioning data from multiple information sources at a smart terminal according to claim 5, characterized in that: The fusion layer constructs a reliable weight matrix: W Where F=7 represents the number of fields to be merged, and S=4 represents the number of information sources; each element in the matrix represents the trust weight value of the s-th information source on the f-th field; Discrete field fusion strategy: For the four discrete label fields of truck operation number, license plate number, container number, and operation status, the fusion result is the output value of the source with the highest credibility weight among all valid sources. Continuous field fusion strategy: For the three continuous numerical fields of location coordinates, driving direction, and vehicle speed, the fusion result is a weighted average of the output values ​​of each valid information source. Timestamp alignment: Based on the unified system clock, the data from each information source are aligned to the same time coordinate system. The timestamp itself does not participate in the weight fusion.

7. The method according to claim 5, characterized in that: The simulation layer uses a motion model to perform finite-time trajectory simulation. When the system detects that the positioning data output by the fusion layer is continuously missing and there are no obstructing road sections or closed areas in front of the operation location in the stack area, the simulation layer is triggered to perform short-range simulation within a finite time window, so that the simulation results are smoothly connected with the known positioning data before and after the missing segment, and finally the output layer outputs a continuous positioning trajectory.

8. The method for fusing and correcting vehicle positioning data from multiple information sources at a smart terminal according to claim 7, characterized in that: The working method of the simulation layer is as follows: First, it detects that there are continuous gaps in the positioning data output by the fusion layer and that there are no obstructing road sections or closed areas in front of the operation location in the stack area; Second, it obtains the last known position p(t0), speed v(t0), and driving direction when the missing area is detected; Third, it obtains the target position of the truck currently operating from the TOS system. ; Step 4, initialize the extrapolation model; Step 5, calculate the guidance acceleration If it is positive, execute Step 6, if it is negative, do not output the extrapolated trajectory. Step six, perform a one-step deduction. Step 7: Triple constraint check. If it passes directly, proceed to step 8; if the speed exceeds the limit, limit the vehicle speed to [specific limit]. Then proceed to step eight; if the passage crosses the boundary, correct the vehicle's lateral position and proceed to step eight; in step eight, determine whether the target position has been reached or the simulation time has exceeded the limit. If so, proceed to step nine; otherwise, return to step five; in step nine, output the simulation trajectory, smoothly connect it with the existing positioning data, and output the complete positioning result.

9. The method for fusing and correcting vehicle positioning data from multiple information sources at a smart terminal according to claim 8, characterized in that: The inference layer uses the following motion model: Wherein, δ is the deduction time step; The guiding acceleration is calculated by the relative relationship between the target position and the current deduction position and the current deduction position ; Initial state: the last known position p(t0), speed v(t0), and direction of travel of the truck when it enters the missing area; Target State: The target berth of the container truck in the TOS system for the current operation ; Constraints: Speed ​​limit in the storage area is 30 km / h (≈8.33 m / s), storage area access width is limited, and parking requirements at the end point are required; A method for fusing and correcting vehicle positioning data from multiple information sources at a smart terminal, as described in claim 9, is characterized in that: Guidance acceleration The calculation of the guidance acceleration is simultaneously subject to three constraints: Terminal constraint: the container truck needs to be stopped precisely under the target berth, i.e. when p(t) → v(t) → 0; Speed constraint: the speed at any time during the deduction process must not exceed the stack area speed limit ; Channel constraints: The stacking area channel is a straight channel, and the lateral offset of the truck is limited by the channel width.

Images