Intelligent regulation and control method for intelligent logistics unmanned metering system
By using multimodal perception and intelligent diagnosis with visual cameras and lidar, the geometric constraints of vehicle parking and the pixel map of cargo loading are generated. Combined with weighing sensors, a parking mechanical consistency criterion is constructed, which solves the problems of vehicle parking effectiveness and anti-cheating in unattended weighing systems, and realizes highly reliable and robust intelligent weighing.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-21
- Publication Date
- 2026-03-13
AI Technical Summary
Existing unattended weighing systems have shortcomings in verifying the validity of vehicle parking, assessing quality reliability, and preventing cheating. They are susceptible to wind loads and vibration interference, cannot identify empty cargo compartments or tarpaulin coverings, and have weak anti-cheating mechanisms, making it difficult to achieve highly reliable and robust intelligent weighing.
By simultaneously acquiring vehicle images and point clouds through visual cameras and LiDAR, a set of geometric constraints for vehicle parking and a pixel map of cargo loading are generated. A parking mechanical consistency criterion is constructed by combining weighing sensors, and visual-assisted quality estimation is integrated to diagnose cheating risks and dynamically select output modes, including direct output, intelligent arbitration, or rejection of measurement.
It achieves fully automated, highly reliable intelligent logistics unmanned weighing, improves weighing accuracy and anti-cheating capabilities, significantly enhances system robustness, and reduces human intervention.
Smart Images
Figure CN121655666A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of smart logistics and industrial automatic weighing technology, specifically to an intelligent control method for a smart logistics unmanned weighing system. Background Technology
[0002] Currently, unattended weighing systems are widely used in closed logistics scenarios such as ports, mines, power plants, and large manufacturing parks. These systems can automatically complete vehicle weighing and data uploading without on-site personnel intervention. These systems typically integrate license plate recognition, weighbridge sensors, and video monitoring modules. They trigger weighing operations through preset procedures and synchronize the results to the business management platform, significantly improving measurement efficiency and reducing labor costs.
[0003] However, existing technologies still have significant shortcomings in ensuring the reliability of weighing and making intelligent decisions: First, the systems generally assume that "the weighing is effective as long as the vehicle is stationary," relying solely on fixed waiting times or weight fluctuation thresholds to determine whether weighing is triggered. They lack active verification of the vehicle's mechanical static state and are easily affected by wind loads, vibrations, or slight swaying. Second, the weighing results rely entirely on the output of a single sensor, without integrating visual or three-dimensional perception information for cross-verification, making it impossible to identify quality misjudgments caused by empty cargo compartments, partial loading, or tarpaulin covering. Third, the anti-cheating mechanism is weak, making it difficult to detect typical cheating behaviors such as "weighing with the edge pressed" and "suspended axle weight" in real time. Most of these methods rely on manual review afterward, failing to achieve process interception and automatic rejection.
[0004] As smart logistics demands increasingly higher levels of measurement accuracy, automation, and anti-fraud capabilities, there is an urgent need for a novel unmanned measurement and control method that integrates multimodal sensing, vehicle dynamics modeling, and intelligent diagnostic decision-making. This method should be able to dynamically assess parking effectiveness, jointly estimate mass reliability, diagnose load anomalies in real time, and adaptively select output strategies accordingly. This would enable a highly reliable, robust, and interpretable intelligent weighing closed loop without any human intervention. Summary of the Invention
[0005] This invention provides an intelligent control method for an unmanned metering system in smart logistics, comprising: S10. Using visual cameras and lidar deployed in the weighbridge area, the exterior images and 3D point clouds of the vehicles entering the weighbridge are collected simultaneously. The geometric constraint set of the vehicle parking and the pixel map of the cargo box are generated by cargo box instance segmentation and ground contact surface detection. S20. Based on the vehicle parking geometric constraint set and combined with the fluctuation characteristics of the time-series output of the weighing sensor, a parking mechanical consistency criterion is constructed. When the criterion meets the preset convergence condition, the vehicle is determined to be in an effective weighing state and the multi-source sensor data of the corresponding time period is latched. S30. Based on the fusion of multi-source sensor data fusion of latching, the geometric volume of the cargo loading matrix and the historical cargo density distribution are used to generate a priori visual-assisted quality estimate, and the residual test is performed with the output of the weighing sensor to obtain the confidence level of measurement consistency. S40. If the confidence level of measurement consistency is lower than the threshold, the abnormal diagnosis sub-process is activated. By comparing the projection of the cargo box center of gravity with the offset of the weighbridge pressure center and analyzing the wheel suspension status and the rationality of the axle load distribution, a cheating risk score and invalid weighing mark are generated. S50. Dynamically select the output mode based on the measurement consistency confidence level and the fraud risk score. When the confidence level is high and the risk score is low, directly output the fusion quality result. When the confidence level is medium and the risk score is medium, trigger high-definition image evidence collection and start the intelligent arbitration process. When the confidence level is low or the risk score is high, reject this measurement and record the abnormal event log.
[0006] The intelligent control method for an unmanned weighing system in smart logistics, as described above, utilizes visual cameras and LiDAR deployed in the weighbridge area to simultaneously acquire exterior images and 3D point clouds of entering vehicles. Through cargo compartment instance segmentation and ground contact surface detection, it generates a vehicle parking geometric constraint set and a cargo compartment loading volume map. Specifically, it includes: The visual image and the laser point cloud are time-stamped and aligned using a hardware synchronization triggering mechanism, and the RANSAC algorithm is used to fit the ground plane to remove point clouds in non-bearing areas. The improved PointPillars network is used to perform 3D bounding box regression of the cargo box, and the tarpaulin-level height interpolation is used to complete the tarpaulin-covered area by combining semantic segmentation mask. The intersection of the completed cargo box-occupied grid and the wheel ground contact convex hull is calculated, and the effective support area ratio is used as the key dimension of the parking geometric constraint.
[0007] The intelligent control method for an unmanned weighing system in smart logistics, as described above, involves constructing a parking mechanical consistency criterion based on the vehicle's parking geometric constraint set and the fluctuation characteristics of the timing output of the weighing sensor. When this criterion meets a preset convergence condition, the vehicle is determined to be in an effective weighing state, and multi-source sensor data for the corresponding time period is latched. Specifically, this includes: The mean of the absolute values of the standard deviation and the first difference within the sliding window are extracted from the output sequence of the weighing sensor to form a dynamic weight stability index. The spatial static reliability is formed by normalizing the variance of the vehicle's attitude angular velocity and the ratio of the effective support area of the cargo box bottom, and then weighting them. The dynamic weight stability index and spatial static reliability are integrated into a stationary mechanics consistency criterion, and an adaptive convergence threshold is set. Data latching is triggered only when the criterion value is higher than the threshold for multiple consecutive periods.
[0008] The intelligent control method for an unmanned metering system in smart logistics, as described above, involves generating a priori visual-assisted mass estimate based on the geometric volume of the cargo container's pixel map, derived from latched multi-source sensor data fusion, and comparing it with historical cargo density distribution. This estimate is then compared with the residuals from the weighing sensor outputs to obtain a confidence level of metering consistency. Specifically, this includes: The effective loading volume is obtained by integrating the three-dimensional voxel map of the cargo compartment along the height direction, and the corresponding typical density range is matched according to the cargo compartment type. By combining the volume-to-weight ratio distribution in historical weighing records of similar vehicles, a Bayesian prior probability density function is constructed. Gaussian residual modeling is performed on the weighing sensor readings and the expected value of the visual volume × density prior, and the posterior probability is calculated as the confidence level of metrological consistency.
[0009] The intelligent control method for an unmanned weighing system in smart logistics, as described above, involves activating an anomaly diagnosis subprocess if the weighing consistency confidence level falls below a threshold. This subprocess compares the cargo box center of gravity projection with the weighbridge pressure center offset and analyzes the wheel suspension status and axle load distribution to generate a cheating risk score and invalid weighing mark. Specifically, this includes: The centroid coordinates are calculated based on the three-dimensional reconstruction results of the cargo compartment, and then projected onto the weighbridge platform to obtain the theoretical pressure center position. By utilizing the output of a multi-sensor array deployed under the weighbridge, the actual pressure center coordinates are inverted through the torque balance equation. If the horizontal distance between the theoretical and actual pressure center exceeds 15% of the cargo box width, or if the weight of any drive bearing is detected to be less than 30% of the expected full load value, then it is determined that there is a cheating behavior of pressing the edge or suspending the load and a high-risk score is generated.
[0010] The intelligent control method for an unmanned metering system in smart logistics, as described above, dynamically selects the output mode based on the metering consistency confidence level and the fraud risk score. When the confidence level is high and the risk score is low, the fusion quality result is directly output. When the confidence level and risk score are moderate, high-definition image evidence collection is triggered and an intelligent arbitration process is initiated. When the confidence level is low or the risk score is high, the metering is rejected and an anomaly event log is recorded. Specifically, this includes: Construct a finite state machine containing three states: "direct output", "intelligent arbitration" and "refusal measurement". The transition conditions are determined by the joint threshold of measurement consistency confidence and fraud risk score. In intelligent arbitration mode, a high-definition PTZ camera is automatically invoked to capture images of the top of the cargo box and the edge area of the weighbridge from multiple angles, and the images are input into the lightweight cheating identification model for secondary judgment. In the case of refused measurement, a structured log containing timestamps, vehicle IDs, exception type codes, and multimodal evidence summaries is generated for subsequent auditing and tracing.
[0011] The intelligent control method for an unmanned weighing system in smart logistics, as described above, involves automatically using a high-definition PTZ camera to capture images of the top of the cargo compartment and the edge area of the weighbridge from multiple angles during intelligent arbitration. These images are then input into a lightweight fraud detection model for secondary judgment. Specifically, this includes: The system automatically plans the pitch and azimuth angles of the PTZ camera based on the 3D reconstruction results of the cargo compartment, ensuring coverage of the opening area at the top of the cargo compartment and the edges of the weighbridge. The PTZ camera is controlled to sequentially perform wide-angle panoramic shooting and close-up shooting, with the close-up area focusing on the joint of the weighbridge platform in the direction of pressure center offset. Digital watermarking and hash signing are performed on the captured images to ensure that the evidence data is tamper-proof, and the data is fed into a lightweight convolutional network based on an attention mechanism to output the final arbitration result.
[0012] This invention also provides an intelligent control system for an unmanned metering system in smart logistics, comprising: The dynamic load semantic field construction module is used to simultaneously acquire the appearance images and 3D point clouds of vehicles entering the weighbridge area using visual cameras and lidar deployed in the weighbridge area. It generates a set of vehicle parking geometric constraints and a pixel map of the cargo loading through cargo box instance segmentation and ground contact surface detection. The weighing trigger decision module is used to construct a parking mechanical consistency criterion based on the vehicle parking geometric constraint set and the fluctuation characteristics of the timing output of the weighing sensor. When the criterion meets the preset convergence condition, it determines that the vehicle is in an effective weighing state and latches the multi-source sensor data of the corresponding time period. The joint quality estimation module is used to generate visual-assisted quality estimation based on the prior generation of geometric volume of cargo loading pixel map and historical cargo density distribution of multi-source sensor data fusion based on latching, and to obtain the measurement consistency confidence degree by performing residual verification with the output of weighing sensor. The measurement anomaly diagnosis module is used to activate the anomaly diagnosis subprocess if the measurement consistency confidence level is lower than the threshold. It generates a cheating risk score and invalid weighing mark by comparing the cargo box center of gravity projection with the weighbridge pressure center offset and analyzing the wheel suspension status and axle load distribution rationality. The intelligent output control module dynamically selects the output mode based on the metrological consistency confidence level and the fraud risk score. When the confidence level is high and the risk score is low, the fusion quality result is directly output. When the confidence level is moderate and the risk score is moderate, high-definition image evidence collection is triggered and the intelligent arbitration process is started. When the confidence level is low or the risk score is high, the current measurement is rejected and the abnormal event log is recorded.
[0013] The beneficial effects achieved by this invention are as follows: This invention provides a fully automatic and highly reliable intelligent logistics unmanned weighing method. It dynamically verifies the effectiveness of vehicle parking through multimodal perception, integrates the three-dimensional volume and density prior of the cargo box to jointly estimate the mass and quantify the confidence level, uses the center of gravity projection and the pressure center offset of the weighbridge to diagnose cheating behaviors such as edge pressing and suspension, and adaptively selects output, arbitration or rejection strategies based on risk scoring. No manual intervention is required throughout the process, which significantly improves the weighing accuracy, anti-cheating ability and system robustness. Attached Figure Description
[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.
[0015] Figure 1 This is a flowchart of an intelligent control method for an unmanned metering system in smart logistics, provided in Embodiment 1 of this application. Figure 2 This is a schematic diagram of an intelligent control system for an unmanned metering system in smart logistics, provided in Embodiment 2 of this application. Detailed Implementation
[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0017] Example 1 like Figure 1 As shown, Embodiment 1 of this application provides an intelligent control method for an unmanned metering system in smart logistics, comprising the following steps: S10. Using visual cameras and lidar deployed in the weighbridge area, the exterior images and 3D point clouds of the vehicles entering the weighbridge are collected simultaneously. The geometric constraint set of the vehicle parking and the pixel map of the cargo box are generated by cargo box instance segmentation and ground contact surface detection. The system strategically deploys industrial-grade global shutter cameras and 16-line or higher solid-state LiDAR sensors around and above the weighbridge to ensure comprehensive coverage of incoming vehicles. All sensors are connected to a unified edge computing unit for real-time data fusion and modeling. The specific steps include: S101. The visual image and the laser point cloud are time-stamped and aligned through a hardware synchronization triggering mechanism, and the RANSAC algorithm is used to fit the ground plane to remove the point cloud in the non-bearing area. To ensure strict spatiotemporal consistency of multi-source sensing data, the system employs a hardware-level triggered synchronization mechanism: the main control module simultaneously sends synchronization pulse signals to the camera and LiDAR, ensuring that both initiate frame acquisition at the same physical moment. Each image frame and its corresponding point cloud are stamped with a nanosecond-level PTP timestamp, with time alignment errors controlled within 1 millisecond. Subsequently, combined with pre-completed internal and external parameter joint calibration results, the point cloud is precisely projected onto the image coordinate system, achieving cross-modal spatial alignment.
[0018] Based on this, the system performs RANSAC plane fitting on the original point cloud to identify the horizontal support plane where the weighbridge platform is located. The maximum number of iterations is set to 500, the inlier distance threshold is 2 cm, and the inlier proportion threshold is 80%. After successful fitting, all points whose Z-axis height deviates from the plane by more than 5 cm are considered non-load-bearing areas (such as weighbridge edges, drainage ditches, road shoulders, or ramps) and are removed, retaining only the point cloud within the effective weighing area to form a clean and reliable set of load-bearing surface point clouds.
[0019] S102. Based on the improved PointPillars network, perform 3D bounding box regression on the cargo box, and combine semantic segmentation mask to perform voxel-level height interpolation to complete the tarpaulin-covered area. The system inputs the preprocessed point cloud into a targeted optimized PointPillars 3D object detection network. Based on the original architecture, this network embeds a channel attention module into the feature extraction backbone to enhance sensitivity to changes in the length, width, and height of the cargo compartment; simultaneously, a lightweight semantic segmentation branch is added after the detection head, outputting the 3D bounding box and its corresponding binary occupancy mask for each detected object. During the training phase, the model uses a large-scale labeled dataset covering various typical logistics vehicle types, including convertibles, standard vans, tarpaulin-covered vans, and curtain-side vans.
[0020] To address the issue of missing point cloud data on the top of the cargo compartment due to tarpaulin covering, the system integrates semantic segmentation results from the visual image. First, the YOLOv7-Seg model performs pixel-level cargo compartment region segmentation on the image, obtaining a cargo compartment contour mask. Then, by calibrating parameters, this contour is back-projected into 3D space, forming a 2D support area for the cargo compartment on the ground. For voxel columns within this area without point cloud coverage (i.e., no effective laser echo in the vertical direction), the system employs a neighborhood-weighted interpolation strategy: using the maximum height of the point cloud in the eight surrounding adjacent voxel columns as a reference, the missing height is calculated using the inverse of the Euclidean distance, thus completing voxel-level incompleteness and ultimately generating a complete cargo compartment occupancy voxel map with a resolution of 0.1m × 0.1m × 0.1m.
[0021] S103. Find the intersection between the completed cargo box-occupied grid and the wheel ground convex hull, and calculate the effective support area ratio as the key dimension of the parking geometric constraint.
[0022] The system integrates visual and point cloud information to accurately locate the contact patch area of each tire. Specifically, it uses wheel detection results from the visual image to obtain the tire center pixel coordinates, and performs cross-validation by combining this with low-density point clusters in the point cloud to improve the robustness of the localization. Based on the preset tire radius in the vehicle model database, a circular contact patch area is generated for each tire, and a convex hull algorithm is performed on all contact patch areas to obtain the overall contact patch convex hull of the vehicle.
[0023] Simultaneously, the projected area of the cargo box's bottom surface on the horizontal plane is extracted from the completed voxel map of the cargo box. A Boolean intersection operation is performed on this projected area and the grounding convex hull, and the ratio of the overlapping area to the total area of the cargo box's bottom surface is calculated; this ratio represents the effective support area ratio. This ratio directly reflects whether the cargo box is completely within the effective load-bearing area of the weighbridge. If this ratio is below 95%, it is preliminarily determined that the vehicle is at risk of being pressed against the edge, partially suspended, or not fully within the weighbridge.
[0024] Finally, the output is a set of vehicle parking geometric constraints that includes a complete voxel map of the cargo box, the proportion of the effective support area, the vehicle attitude angle, and the coordinates of the ground convex hull boundary.
[0025] S20. Based on the vehicle parking geometric constraint set and combined with the fluctuation characteristics of the time-series output of the weighing sensor, a parking mechanical consistency criterion is constructed. When the criterion meets the preset convergence condition, the vehicle is determined to be in an effective weighing state and the multi-source sensor data of the corresponding time period is latched. This step aims to identify the weighing window where the vehicle is truly stationary and the load is stable from dynamic sensing signals, avoiding measurement errors caused by the vehicle not being fully stopped, minor vibrations, or human interference. The system integrates the temporal stability of weight with the rationality of spatial stationary position to construct a unified criterion, and introduces an adaptive threshold mechanism to improve robustness.
[0026] S201. Extract the mean of the absolute values of the standard deviation and the first-order difference within the sliding window from the output sequence of the weighing sensor to form a dynamic weight stability index. The system collects outputs from various sensors on the weighbridge at a frequency of 10 Hz and uses a 3-second sliding window to calculate two core statistics in real time: the standard deviation of the total weight readings within the window, reflecting the instantaneous fluctuation amplitude; and the mean of the absolute values of the first-order differences, characterizing the smoothness of the weight change trend. These two values are normalized and weighted together to form a dynamic weight stability index. This index effectively distinguishes between "true stability" and "pseudo-stability" (such as slow loading, driver not turning off the engine, etc.) and is the primary basis for determining the validity of the weighing.
[0027] S202. After normalizing the variance of the vehicle's attitude angular velocity and the ratio of the effective support area of the cargo box bottom, a weighted average is taken to form the spatial static reliability. The short-time variance is calculated using the angular velocity data provided by the IMU. Combined with the effective support area ratio obtained in S10, the variances are normalized and weighted to generate spatial static confidence. This is used to exclude situations where the vehicle is stable in weight but has geometric anomalies such as tilting, being suspended, or not fully driven into the vehicle.
[0028] S203. The dynamic weight stability index and spatial static reliability are integrated into a stationary mechanical consistency criterion, and an adaptive convergence threshold is set. Data latching is triggered only when the criterion value is higher than the threshold for multiple consecutive periods.
[0029] The system integrates the aforementioned two indicators into a unified criterion and introduces an adaptive convergence threshold mechanism: the initial threshold is set to 0.85. If the criterion value exceeds the threshold for three consecutive 1-second cycles, data latching is triggered; otherwise, the threshold is dynamically adjusted based on the criterion distribution of historical successful weighing samples to prevent missed or false judgments in interference scenarios such as strong winds and ground vibrations. Once the condition is met, the system immediately latches all aligned multi-source sensor data within that time period.
[0030] S30. Based on the fusion of multi-source sensor data fusion of latching, the geometric volume of the cargo loading matrix and the historical cargo density distribution are used to generate a priori visual-assisted quality estimate, and the residual test is performed with the output of the weighing sensor to obtain the confidence level of measurement consistency. S301. Integrate the three-dimensional voxel map of the cargo box along the height direction to obtain the effective loading volume, and match the corresponding typical density range according to the cargo box type; The system generates a voxel map of the cargo compartment occupancy in S10. Integrate along the Z-axis, count the number of effective voxels at each horizontal grid point, multiply by the volume of a single voxel, and sum to obtain the total loading volume. Meanwhile, based on the cargo category labels (such as "high-sided truck", "box truck", "tanker truck" etc.) output by the PointPillars network in S10, the historical typical cargo density range corresponding to this type is retrieved from the preset knowledge base as the semantic constraint boundary for subsequent prior modeling to prevent unreasonable density extrapolation.
[0031] S302. Construct a Bayesian prior probability density function by combining the volume-weight ratio distribution in historical weighing records of similar vehicles. The system retrieves all weighing records from the cloud-based historical database that match the current vehicle's cargo compartment type. A total of N valid historical samples are obtained, each record containing the visually reconstructed volume. With corresponding weighing mass And calculate the load-bearing ratio per unit volume. This ratio reflects the equivalent density of goods actually loaded under a specific cargo compartment type. For this sample set... A non-parametric density ratio prior is constructed using Gaussian kernel density estimation: ,in, For cargo container type, For adaptive bandwidth. Therefore, the prior distribution of the expected quality of the current vehicle is: This enables a probability mapping from historical statistics to the current scenario.
[0032] S303. Gaussian residual modeling is performed on the weighing sensor readings and the expected value of the visual volume × density prior, and the posterior probability is calculated as the confidence level of metrological consistency.
[0033] Let the average reading within the latching window of the weighing sensor be... Its measurement noise standard deviation is The system will With prior distribution Perform Bayesian fusion, defining the econometric consistency confidence score as the normalized marginal likelihood of the measured value supported by historical priors: The numerator of this formula represents the probability of the current weighing reading occurring under a historically reasonable density distribution, while the denominator is the theoretically maximum possible response, ensuring... .when When the load falls within a reasonable range supported by historical data, the confidence level is close to 1; if it deviates significantly (such as an empty vehicle falsely reporting a heavy load, or a fully loaded vehicle with abnormally low quality), the confidence level decays rapidly. This confidence level will be directly used for the S40's anomaly trigger judgment and the S50's output mode selection.
[0034] S40. If the confidence level of measurement consistency is lower than the threshold, the abnormal diagnosis sub-process is activated. By comparing the projection of the cargo box center of gravity with the offset of the weighbridge pressure center and analyzing the wheel suspension status and the rationality of the axle load distribution, a cheating risk score and invalid weighing mark are generated. when When the weight falls below a preset threshold, the system determines that the current weighing result poses an abnormal risk and automatically triggers this abnormality diagnosis sub-process. This process cross-verifies the vehicle's stationary state from two dimensions: geometric mechanics and load distribution. It focuses on identifying typical cheating behaviors such as "edge pressing," "single-sided suspension," and "axle load transfer," and outputs a structured cheating risk score and invalid weighing mark. Specifically, it includes the following sub-steps: S401. Calculate the centroid coordinates based on the three-dimensional reconstruction results of the cargo box and project them onto the weighbridge platform to obtain the theoretical pressure center position. The system utilizes the complete cargo compartment occupancy voxel map generated by S10. Assuming the cargo is uniformly distributed, calculate the geometric centroid of the loaded portion of the cargo compartment. Specifically, for the coordinates of the center of all valid voxels... Calculate the arithmetic mean to obtain the three-dimensional centroid. Since the weighbridge only responds to vertical loads, projecting this point vertically onto the weighbridge platform yields the theoretical center of pressure. This point represents the location where the resultant force of the entire vehicle should be applied under ideal static load and without cheating.
[0035] S402. Using the output of a multi-sensor array deployed under the weighbridge, the actual pressure center coordinates are inverted through the torque balance equation. The weighbridge is equipped with multiple independent weighing sensors at its bottom, and the system reads the real-time output of each sensor. and its installation coordinates According to the principle of static torque balance, the total torque of the vehicle's gravity about the origin is equal to the sum of the torques of all the reaction forces. From this, the actual center of pressure can be determined. : , Where M represents the number of effective sensors. This coordinate reflects the equivalent point of application of the actual load on the weighbridge and is extremely sensitive to off-center loading and suspension.
[0036] S403. If the horizontal distance between the theoretical and actual pressure center exceeds 15% of the cargo box width, or if the weight of any drive bearing is detected to be less than 30% of the expected full load value, then it is determined that there is a cheating behavior of pressing the edge or suspending the air and a high-risk score is generated.
[0037] The system first calculates the Euclidean distance between the theoretical and actual pressure centers. and the width of the cargo box Normalized comparisons are performed. Simultaneously, based on the vehicle axle type identification results, the measured axle load is obtained for each drive axle k. (Derived from sensor array decomposition) and expected axle load at full load (Estimated based on historical axle load distribution or static lever model of similar vehicles).
[0038] Based on this, a cheating risk scoring function that integrates geometric offset and axle load anomaly is proposed, introducing nonlinear decay and logic gating mechanisms to avoid misjudgment by a single indicator: ,in, This is the geometric offset sensitivity coefficient, which controls the rate of risk increase after the offset exceeds the limit; For the set of drive shafts; For the Sigmoid function, The parameter represents the axle load abnormal steepness; the first parameter measures the risk of pressure center shift: when d ≤ When the first item is 0, it grows exponentially to close to 1 after exceeding the limit; the second item measures the risk of axle load being suspended: as long as the measured load ratio of any drive axle is... 30% below the expected full load corresponds to a Sigmoid output close to 1, a product term close to 0, and an overall second term close to 1; ultimately The higher the value, the higher the probability of cheating.
[0039] like The system generates a high-risk cheating score and marks the weighing as invalid. It also records diagnostic information such as offset direction and suspended axis position, and automatically intercepts the weighing. This mechanism effectively distinguishes between normal slight deviations and malicious edge / suspended behavior, significantly improving the anti-cheating capabilities and judicial credibility of unmanned weighing systems.
[0040] S50. Dynamically select the output mode based on the measurement consistency confidence level and the fraud risk score. When the confidence level is high and the risk score is low, directly output the fusion quality result. When the confidence level is medium and the risk score is medium, trigger high-definition image evidence collection and start the intelligent arbitration process. When the confidence level is low or the risk score is high, reject this measurement and record the abnormal event log.
[0041] S501. Construct a finite state machine containing three states: "direct output", "intelligent arbitration" and "rejection of measurement". Its transition conditions are determined by the joint threshold of measurement consistency confidence and fraud risk score. When the measurement consistency confidence level is high and the cheating risk score is low, it indicates that the weighing result is highly reliable and no signs of cheating have been detected. At this time, the system will directly output the fused quality results, including information such as vehicle gross weight and axle load distribution.
[0042] If the confidence level for metrological consistency and the cheating risk score are at a moderate level, it indicates that although there may be some uncertainty in the weighing results, it is not enough to completely negate their validity. In this case, the system will initiate an intelligent arbitration process. In this state, the system will automatically call upon high-definition monitoring equipment to collect evidence and use a lightweight model for secondary analysis to confirm whether cheating has occurred.
[0043] If the measurement consistency confidence level is low or the cheating risk score is high, it means that there is a significant anomaly or strong suspicion of cheating during the weighing process. At this time, the system will not only refuse to accept the current weighing result, but also generate a detailed anomaly event log for subsequent auditing.
[0044] S502: In intelligent arbitration mode, the high-definition PTZ camera is automatically called to capture images of the top of the cargo box and the edge area of the weighbridge from multiple angles, and the images are input into the lightweight cheating identification model for secondary judgment. Upon entering intelligent arbitration mode, the system first plans the optimal shooting angle for the high-definition PTZ camera based on the results of the 3D reconstruction of the cargo compartment, ensuring full coverage of the opening area at the top of the cargo compartment and the edges of the weighbridge. The specific operations are as follows: Based on the actual size and location of the cargo compartment, the system calculates the optimal pitch and azimuth angles that the PTZ camera's pan / tilt head needs to be adjusted to; it controls the PTZ camera to execute a series of preset actions, including wide-angle panoramic shooting to obtain an overall view, and close-up shooting of the weighbridge platform seams in the direction of pressure center offset to capture potential cheating details; all captured images are digitally watermarked and hash-signed to ensure the authenticity and integrity of the evidence data are not tampered with.
[0045] Ultimately, these images are fed into a lightweight convolutional network based on an attention mechanism, and after analysis by a deep learning model, a final arbitration conclusion is reached.
[0046] S503. When metering is refused, generate a structured log containing timestamps, vehicle IDs, exception type codes, and multimodal evidence summaries for subsequent auditing and tracing.
[0047] When the system decides to reject a measurement, it automatically generates a structured log entry. This log contains key information elements such as a timestamp, vehicle ID, exception type code, and a multimodal evidence summary. This design facilitates future event tracking and investigation, and also provides valuable data support for preventing similar problems from occurring. Furthermore, recording a detailed evidence summary effectively improves the transparency and fairness of the entire system.
[0048] Example 2 like Figure 2 As shown, Embodiment 2 of this application provides an intelligent control system for an unmanned metering system in smart logistics, comprising: The dynamic load semantic field construction module utilizes visual cameras and LiDAR deployed in the weighbridge area to simultaneously acquire exterior images and 3D point clouds of entering vehicles. It then generates a vehicle parking geometric constraint set and a cargo loading pixel map through cargo box instance segmentation and ground contact surface detection. Specifically, it is divided into the following sub-modules: Multi-source sensing synchronization submodule: It realizes frame-level time alignment between industrial-grade global shutter camera and solid-state LiDAR through hardware level triggering mechanism, and completes cross-modal spatial registration of point cloud and image based on joint calibration parameters to ensure spatiotemporal consistency of multi-source data.
[0049] The load-bearing surface purification and cargo box reconstruction submodule uses the RANSAC algorithm to fit the horizontal support plane of the weighbridge platform and removes the point cloud of non-load-bearing areas with height deviation. At the same time, it performs 3D bounding box regression of the cargo box based on the improved PointPillars network and performs voxel-level height interpolation on the tarpaulin-covered area by integrating visual semantic segmentation mask to generate a complete voxel map of cargo box occupancy.
[0050] The parking geometric constraint generation submodule combines the convex hull of the wheel grounding area and the projection area of the cargo box bottom surface, calculates the ratio of the intersection area of the two to the total area of the cargo box bottom surface as the effective support area ratio, and outputs a structured geometric constraint set including attitude angles, grounding boundaries and voxel maps.
[0051] The weighing trigger decision module is used to construct a parking mechanical consistency criterion based on the vehicle's parking geometric constraint set and the fluctuation characteristics of the timing output of the weighing sensor. When the criterion meets the preset convergence condition, the vehicle is determined to be in a valid weighing state, and the multi-source sensor data for the corresponding time period is latched. Specifically, it is divided into the following sub-modules: Weight stability assessment submodule: Extracts the mean of the standard deviation and the first-order difference absolute value within the sliding window from the output sequence of the weighing sensor, and forms a dynamic weight stability index after normalization and weighting, which characterizes the instantaneous fluctuation and trend smoothness of the load.
[0052] Spatial Static Credibility Submodule: It integrates the angular velocity variance output by the inertial measurement unit with the effective support area ratio, normalizes them separately, and then weights them to generate spatial static credibility, which is used to exclude geometric anomalies such as vehicle tilting, suspension, or not fully entering the vehicle.
[0053] The adaptive latching control submodule integrates weight stability and spatial static reliability into a unified criterion for stationary mechanics consistency. It introduces a continuous periodic achievement mechanism and a threshold adaptive strategy driven by historical samples, which triggers multi-source sensor data latching when the convergence condition is met.
[0054] The joint quality estimation module is used to generate a priori visual-assisted quality estimate based on the geometric volume of the cargo container's pixel map and the historical cargo density distribution, fused from latched multi-source sensor data. It then performs residual verification with the weighing sensor output to obtain the measurement consistency confidence level. Specifically, it is divided into the following sub-modules: The geometric volume and type recognition submodule integrates the voxel map of the cargo compartment along the height direction to obtain the effective loading volume, and matches the corresponding typical cargo density range based on the cargo compartment category label output by the 3D detection.
[0055] Historical density prior modeling submodule: retrieves records of similar vehicles from the cloud-based historical weighing database, extracts samples of load-to-weight ratio per unit volume, and uses nonparametric kernel density estimation to construct the probability distribution of this ratio as the density prior.
[0056] The consistency confidence assessment submodule models the probability residuals between the average reading of the weighing sensor and the expected mass distribution generated based on priors, and calculates the normalized posterior response as the metrological consistency confidence, which is used to quantify the credibility of the weighing results.
[0057] The measurement anomaly diagnosis module is used to activate the anomaly diagnosis sub-process if the measurement consistency confidence level is lower than a threshold. It generates a cheating risk score and invalid weighing mark by comparing the cargo box center of gravity projection with the weighbridge pressure center offset and analyzing the wheel suspension status and axle load distribution rationality. Specifically, it is divided into the following sub-modules: Theoretical pressure center calculation submodule: Calculates the geometric centroid of the loaded part based on the voxel map of the cargo box, and projects it vertically onto the weighbridge platform to obtain the position of the theoretical pressure center.
[0058] Actual pressure center inversion submodule: Utilizing the output of the multi-point weighing sensor array at the bottom of the weighbridge and its installation coordinates, the actual pressure center coordinates of the entire vehicle are inverted through the torque balance principle.
[0059] The comprehensive risk assessment submodule for cheating integrates the offset distance of the pressure center with the deviation of the measured load on the drive shaft from the expected value of full load, and constructs a nonlinear risk scoring mechanism. When the score exceeds the set threshold, it is marked as invalid weighing and the cheating characteristics are recorded.
[0060] The intelligent output control module dynamically selects the output mode based on the metrological consistency confidence level and the fraud risk score. When the confidence level is high and the risk score is low, it directly outputs the fusion quality result. When the confidence level and risk score are moderate, it triggers high-definition image forensics and initiates the intelligent arbitration process. When the confidence level is low or the risk score is high, it rejects the current measurement and records the abnormal event log. Specifically, it is divided into the following sub-modules: State Decision Submodule: Constructs a finite state machine containing three states: "direct output", "intelligent arbitration" and "reject measurement". Its state transition is dynamically determined by the joint threshold of measurement consistency confidence and fraud risk score.
[0061] High-definition evidence collection and execution submodule: In intelligent arbitration mode, based on the 3D reconstruction results of the cargo compartment, the tilt and azimuth angles of the PTZ camera are automatically planned, wide-angle panoramic and weighbridge edge close-up shooting are performed, digital watermarks and hash signatures are embedded in the images, and the images are input into a lightweight cheating identification model for secondary judgment.
[0062] The anomaly log generation submodule automatically generates structured audit logs containing timestamps, vehicle identification, anomaly type codes, and multimodal evidence summaries when metering is refused, supporting post-event traceability and system optimization.
[0063] Corresponding to the above embodiments, the present invention provides a computer storage medium, including: at least one memory and at least one processor; The memory is used to store one or more program instructions; A processor is used to run one or more program instructions to execute an intelligent control method for an unmanned metering system in smart logistics.
[0064] Corresponding to the above embodiments, this embodiment of the invention provides a computer-readable storage medium containing one or more program instructions, which are executed by a processor to provide an intelligent control method for an unmanned metering system in smart logistics.
[0065] The embodiments disclosed in this invention provide a computer-readable storage medium storing computer program instructions. When the computer program instructions are executed on a computer, the computer performs the above-described intelligent control method for an unmanned metering system for smart logistics.
[0066] In this embodiment of the invention, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0067] The various methods, steps, and logic diagrams disclosed in the embodiments of this invention can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor reads information from the storage medium and, in conjunction with its hardware, completes the steps of the above methods.
[0068] The storage medium can be memory, such as volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.
[0069] Among them, non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory.
[0070] Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (Synchlink DRAM, SLDRAM), and direct memory bus RAM (DRRAM).
[0071] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.
[0072] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using a combination of hardware and software. When applied as software, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of computer programs from one place to another. Storage media can be any available medium that can be accessed by a general-purpose or special-purpose computer.
[0073] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.
Claims
1. An intelligent control method for an unmanned metering system in smart logistics, characterized in that, Includes the following steps: S10. Using visual cameras and lidar deployed in the weighbridge area, the exterior images and 3D point clouds of the vehicles entering the weighbridge are collected simultaneously. The geometric constraint set of the vehicle parking and the pixel map of the cargo box are generated by cargo box instance segmentation and ground contact surface detection. S20. Based on the vehicle parking geometric constraint set and combined with the fluctuation characteristics of the time-series output of the weighing sensor, a parking mechanical consistency criterion is constructed. When the criterion meets the preset convergence condition, the vehicle is determined to be in an effective weighing state and the multi-source sensor data of the corresponding time period is latched. S30. Based on the fusion of multi-source sensor data fusion of latching, the geometric volume of the cargo loading matrix and the historical cargo density distribution are used to generate a priori visual-assisted quality estimate, and the residual test is performed with the output of the weighing sensor to obtain the confidence level of measurement consistency. S40. If the confidence level of measurement consistency is lower than the threshold, the abnormal diagnosis sub-process is activated. By comparing the projection of the cargo box center of gravity with the offset of the weighbridge pressure center and analyzing the wheel suspension status and the rationality of the axle load distribution, a cheating risk score and invalid weighing mark are generated. S50. Dynamically select the output mode based on the measurement consistency confidence level and the fraud risk score. When the confidence level is high and the risk score is low, directly output the fusion quality result. When the confidence level is medium and the risk score is medium, trigger high-definition image evidence collection and start the intelligent arbitration process. When the confidence level is low or the risk score is high, reject this measurement and record the abnormal event log.
2. The intelligent control method for an unmanned metering system in smart logistics according to claim 1, characterized in that, The system utilizes visual cameras and LiDAR deployed in the weighbridge area to simultaneously acquire exterior images and 3D point clouds of entering vehicles. Through cargo box instance segmentation and ground contact surface detection, it generates a vehicle parking geometric constraint set and a cargo box loading volume map. Specifically, it consists of the following sub-steps: The visual image and the laser point cloud are time-stamped and aligned using a hardware synchronization triggering mechanism, and the RANSAC algorithm is used to fit the ground plane to remove point clouds in non-bearing areas. The improved PointPillars network is used to perform 3D bounding box regression of the cargo box, and the tarpaulin-level height interpolation is used to complete the tarpaulin-covered area by combining semantic segmentation mask. The intersection of the completed cargo box-occupied grid and the wheel ground contact convex hull is calculated, and the effective support area ratio is used as the key dimension of the parking geometric constraint.
3. The intelligent control method for an unmanned metering system in smart logistics according to claim 1, characterized in that, Based on the vehicle's stationary geometric constraint set and the fluctuation characteristics of the time-series output of the weighing sensor, a stationary mechanical consistency criterion is constructed. When this criterion meets the preset convergence condition, the vehicle is determined to be in an effective weighing state, and the multi-source sensor data for the corresponding time period is latched. Specifically, the process consists of the following sub-steps: The mean of the absolute values of the standard deviation and the first difference within the sliding window are extracted from the output sequence of the weighing sensor to form a dynamic weight stability index. The spatial static reliability is formed by normalizing the variance of the vehicle's attitude angular velocity and the ratio of the effective support area of the cargo box bottom, and then weighting them. The dynamic weight stability index and spatial static reliability are integrated into a stationary mechanics consistency criterion, and an adaptive convergence threshold is set. Data latching is triggered only when the criterion value is higher than the threshold for multiple consecutive periods.
4. The intelligent control method for an unmanned metering system in smart logistics according to claim 1, characterized in that, Based on the fusion of latched multi-source sensor data, the geometric volume of the cargo container's pixel map and the historical cargo density distribution are used to generate a priori visual-assisted quality estimate. This estimate is then compared with the weighing sensor outputs to perform a residual check, yielding a confidence level of measurement consistency. The specific steps are as follows: The effective loading volume is obtained by integrating the three-dimensional voxel map of the cargo compartment along the height direction, and the corresponding typical density range is matched according to the cargo compartment type. By combining the volume-to-weight ratio distribution of historical weighing records of similar vehicles, a Bayesian prior probability density function is constructed. Gaussian residual modeling is performed on the weighing sensor readings and the expected value of the visual volume × density prior, and the posterior probability is calculated as the confidence level of metrological consistency.
5. The intelligent control method for an unmanned metering system in smart logistics according to claim 1, characterized in that, If the confidence level of measurement consistency is lower than the threshold, the anomaly diagnosis subprocess is activated. This subprocess compares the cargo box center of gravity projection with the weighbridge pressure center offset and analyzes the wheel suspension status and axle load distribution to generate a cheating risk score and invalid weighing mark. Specifically, it consists of the following sub-steps: The centroid coordinates are calculated based on the three-dimensional reconstruction results of the cargo compartment, and then projected onto the weighbridge platform to obtain the theoretical pressure center position. By utilizing the output of a multi-sensor array deployed under the weighbridge, the actual pressure center coordinates are inverted through the torque balance equation. If the horizontal distance between the theoretical and actual pressure center exceeds 15% of the cargo box width, or if the weight of any drive bearing is detected to be less than 30% of the expected full load value, then it is determined that there is a cheating behavior of pressing the edge or suspending the load and a high-risk score is generated.
6. The intelligent control method for an unmanned metering system in smart logistics according to claim 1, characterized in that, The output mode is dynamically selected based on the metrological consistency confidence level and the fraud risk score. When the confidence level is high and the risk score is low, the fusion quality result is directly output. When the confidence level and the risk score are moderate, high-definition image forensics are triggered and the intelligent arbitration process is initiated. When the confidence level is low or the risk score is high, the current measurement is rejected and the abnormal event log is recorded. Specifically, the process is divided into the following sub-steps: Construct a finite state machine containing three states: "direct output", "intelligent arbitration" and "reject measurement". Its transition conditions are determined by the joint threshold of measurement consistency confidence and fraud risk score. In intelligent arbitration mode, a high-definition PTZ camera is automatically invoked to capture images of the top of the cargo box and the edge area of the weighbridge from multiple angles, and the images are input into the lightweight cheating identification model for secondary judgment. In the case of refused measurement, a structured log containing timestamps, vehicle IDs, exception type codes, and multimodal evidence summaries is generated for subsequent auditing and tracing.
7. The intelligent control method for an unmanned metering system in smart logistics according to claim 6, characterized in that, In intelligent arbitration mode, a high-definition PTZ camera is automatically invoked to capture images of the top of the cargo compartment and the edge area of the weighbridge from multiple angles, and the images are input into the lightweight cheating detection model for secondary judgment. Specifically, this involves the following sub-steps: The system automatically plans the pitch and azimuth angles of the PTZ camera based on the 3D reconstruction results of the cargo compartment, ensuring coverage of the opening area at the top of the cargo compartment and the edges of the weighbridge. The PTZ camera is controlled to sequentially perform wide-angle panoramic shooting and close-up shooting, with the close-up area focusing on the joint of the weighbridge platform in the direction of pressure center offset. Digital watermarking and hash signing are performed on the captured images to ensure that the evidence data is tamper-proof, and the data is fed into a lightweight convolutional network based on an attention mechanism to output the final arbitration result.
8. An intelligent control system for an unmanned metering system in smart logistics, characterized in that, include: The dynamic load semantic field construction module is used to simultaneously acquire the appearance images and 3D point clouds of vehicles entering the weighbridge area using visual cameras and lidar deployed in the weighbridge area. It generates a set of vehicle parking geometric constraints and a pixel map of the cargo loading through cargo box instance segmentation and ground contact surface detection. The weighing trigger decision module is used to construct a parking mechanical consistency criterion based on the vehicle parking geometric constraint set and the fluctuation characteristics of the timing output of the weighing sensor. When the criterion meets the preset convergence condition, it determines that the vehicle is in an effective weighing state and latches the multi-source sensor data of the corresponding time period. The joint quality estimation module is used to generate visual-assisted quality estimation based on the prior generation of geometric volume of cargo loading pixel map and historical cargo density distribution of multi-source sensor data fusion based on latching, and to obtain the measurement consistency confidence degree by performing residual verification with the output of weighing sensor. The measurement anomaly diagnosis module is used to activate the anomaly diagnosis subprocess if the measurement consistency confidence level is lower than the threshold. It generates a cheating risk score and invalid weighing mark by comparing the cargo box center of gravity projection with the weighbridge pressure center offset and analyzing the wheel suspension status and axle load distribution rationality. The intelligent output control module dynamically selects the output mode based on the metrological consistency confidence level and the fraud risk score. When the confidence level is high and the risk score is low, the fusion quality result is directly output. When the confidence level is moderate and the risk score is moderate, high-definition image evidence collection is triggered and the intelligent arbitration process is started. When the confidence level is low or the risk score is high, the current measurement is rejected and the abnormal event log is recorded.