Container truck anti-hoisting method based on multi-technology fusion

Abstract

The invention provides a container truck anti-hoisting method based on multi-technology fusion. The container truck anti-hoisting method comprises the following steps of visual anti-hoisting, wherein an image is obtained from a camera at the bottom of a large machine leg in real time; image recognition; tracking a tire motion state; judging the horizontal movement state of the tire; judging the vertical movement state of the tire; laser radar hoisting prevention: establishing a shore bridge body coordinate system; after obtaining three-dimensional point cloud data, carrying out preprocessing operation; dynamically constructing a ground reference; dividing a monitoring layer, and judging whether the container truck is hoisted or not by analyzing point cloud data of the monitoring layer in real time; and performing anti-interference processing. The method has the beneficial effects that high-precision judgment on the lifting state of the container truck is realized, the accuracy rate reaches 99.7%, the omission ratio is lower than 0.05%, the false drop rate is lower than 0.1%, and reliable safety guarantee is provided for automatic loading and unloading operation of a wharf.

Description

Technical Field

[0001] This invention belongs to the field of port loading and unloading technology, and in particular relates to a method for preventing container trucks from being lifted based on the integration of multiple technologies. Background Technology

[0002] With the rapid development of global trade, the container throughput of ports, terminals, and other logistics hubs is constantly increasing. To improve operational efficiency and reduce labor costs, automated equipment such as automated rail-mounted cranes (ARMG) has been widely used in container loading and unloading operations. However, while these devices bring high-efficiency operations, they also face many safety hazards, among which the accidental lifting of trucks is a safety issue that urgently needs to be addressed.

[0003] In container handling operations, if the truck's locking pins are not fully released or are not properly released, the crane spreader may lift both the container and the truck together, leading to serious safety accidents. Such accidents not only cause property damage but can also endanger the lives of operators. The core of port automation and safe production lies in addressing the safety hazards and efficiency bottlenecks inherent in traditional container handling operations. Early container handling relied on manual judgment of the locking pin status; truck drivers or remote operators were prone to negligence, resulting in the locking pins not being fully released, causing the spreader to lift the truck along with the container. Therefore, developing an effective truck anti-lifting system is particularly important. Summary of the Invention

[0004] In view of this, the present invention aims to propose a method for preventing container trucks from being lifted based on the integration of multiple technologies, so as to solve the safety hazards in traditional container loading and unloading operations.

[0005] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0006] A method for preventing container trucks from being lifted based on multi-technology integration includes the following steps:

[0007] S1. Visual anti-lifting feature;

[0008] S2, LiDAR anti-lifting;

[0009] In step S1, visual anti-lifting includes:

[0010] S11. Real-time image acquisition from the camera at the bottom of the large leg;

[0011] S12, Image Recognition: Number the identified tires and record their location information;

[0012] S13. By analyzing the number and location information of each tire, the movement status of each tire can be tracked.

[0013] S14. Determine if the tires are in a horizontal movement state. If so, it means that the truck is in a transportation state, rather than in a loading and unloading working state.

[0014] S15. If the tire is in a horizontal stationary state, the vertical movement state is judged. Furthermore, if the tire continues to move upward in the vertical direction, an alarm signal and a stop command are issued.

[0015] In step S2, the lidar anti-lifting mechanism includes:

[0016] S21. Establish the coordinate system of the quay crane body;

[0017] S22. After acquiring the 3D point cloud data, perform preprocessing operations;

[0018] S23, Dynamic construction of ground benchmarks;

[0019] S24. Divide the monitoring layers and determine whether the truck has been lifted by real-time analysis of the point cloud data of the monitoring layers.

[0020] S25, Anti-interference processing;

[0021] In step S1, the visual anti-lifting algorithm uses the intelligent recognition algorithm YOLOv5 and the ByteTrack algorithm.

[0022] Furthermore, in step S21, the coordinate system of the quay crane body is established, including:

[0023] The origin O is the orthogonal projection point of the center line of the quay crane track onto the horizontal plane. The Z-axis is defined vertically, the Y-axis is defined parallel to the track, the positive direction of the Y-axis is the direction of the quay crane's movement, and the X-axis is defined perpendicular to the track, with the positive direction of the X-axis pointing towards the sea.

[0024] Furthermore, in step S22, after acquiring the 3D point cloud data, preprocessing operations are performed, including:

[0025] S221. Remove noise points with a reflection intensity lower than 500 by intensity filtering;

[0026] S222, Distance Truncation: Preserve the effective region X∈[-8.2m,8.2m], Y∈[0.5m,30.5m], Z∈[0,0.8m];

[0027] S223. Compensate for the vibration of the quay crane structure at a frequency of 200Hz using the radar's built-in IMU data.

[0028] Furthermore, in step S23, the dynamic construction of the ground reference includes:

[0029] S231. Using the improved RANSAC plane fitting algorithm, 1000 points are randomly sampled in the interval Z∈[0,15cm], the plane equation is calculated and iterated 1000 times. The expression of the plane equation is as follows:

[0030] ax + by + cz + d = 0;

[0031] S232. Select a plane with more than 500 internal points and a residual of less than 3cm as the ground reference.

[0032] S233. The position of the datum plane is determined by calculating the height of the datum plane. The calculation expression is as follows:

[0033] Z_base=(-d-aX_avg-bY_avg) / c;

[0034] In the formula, X_avg and Y_avg are the mean values ​​of the interior points.

[0035] Furthermore, in step S24, the point cloud data of the monitoring layer is analyzed in real time, including:

[0036] S241. Divide the space 0-80cm above ground into 16 monitoring layers with a height of 5cm, and each monitoring layer covers a 30m horizontal span.

[0037] S242. Maintain a density matrix Di[j][k] for each monitoring layer i, where j∈[0,32] represents the X-axis grid index and k∈[0,59] represents the Y-axis grid index;

[0038] S243. For each point P(x,y,z), calculate the relative height using the following expression:

[0039] Δz = z - Z_base;

[0040] The layer number is calculated using the following expression:

[0041] i = floor(Δz / 0.05);

[0042] The grid coordinates are calculated using the following expression:

[0043] j = floor((x + 8.2) / 0.5);

[0044] k = floor((y-0.5) / 0.5);

[0045] Based on this coordinate mapping, the number of valid points N(i,j,k) in each grid cell (i,j,k) is recorded, and the density value is defined as follows:

[0046] D(i,j,k) = N(i,j,k) / V;

[0047] In the formula, V is the volume of the grid cell, which is 0.5m × 0.5m × 0.05m.

[0048] S243, Basic layer point cloud density monitoring;

[0049] S244, Multilayer density gradient analysis;

[0050] S245. Calculate the density gradient between adjacent layers;

[0051] S246. Introducing time window analysis, when the pattern described in steps S243 to S245 is observed for T consecutive frames, it is confirmed that the truck has been lifted. The lifting probability expression is as follows:

[0052] P=(1 / T)∑ i＝1 T I(ΔD0<-δ0&&G_pattern_match);

[0053] In the formula, I is an indicator function; if P > 0.8, the truck is lifted.

[0054] S247. Examine the spatial continuity of point cloud density changes, set a minimum connected region S_min, and if the continuously changing region S>S_min, it is determined to be a valid lifting event;

[0055] S248. The change in point cloud height distribution entropy is expressed as follows:

[0056] ΔH = H_current - H_base;

[0057] The change in the reflection intensity spectrum is expressed as follows:

[0058] ΔI=||I_current-I_base||2;

[0059] Connected region shape characteristics: Calculate the matching degree between the minimum bounding rectangle and the standard dimensions of the truck;

[0060] Feature-weighted fusion, the expression is as follows:

[0061] Score=w1·ΔD0+w2·G_pattern+w3·ΔH+w4·ΔI+w5·Shape_match;

[0062] In the formula, if Score > τ, then the lifting state is confirmed.

[0063] Furthermore, in step S243, the base layer point cloud density monitoring includes:

[0064] Base layer: i = 0, Z = 0-5cm;

[0065] The point cloud density D(0,j,k) of the base layer is monitored, and the expression is as follows:

[0066] ΔD0=D_base(0,j,k)-D_current(0,j,k)<-δ0;

[0067] In the formula, δ0 is the threshold;

[0068] If the truck is lifted, the density of the base layer point cloud will decrease compared to the baseline value.

[0069] Furthermore, in step S244, the multilayer density gradient analysis includes:

[0070] If the truck is lifted, the point cloud distribution of all layers will shift upwards as a whole.

[0071] Further, in step S245, the density gradient between adjacent layers is calculated, including:

[0072] The density gradient expression between adjacent layers is as follows:

[0073] G(i) = D(i+1,j,k) - D(i,j,k);

[0074] If the container truck is lifted, the density gradient sequence {G(0),G(1),...,G(14)} will show a decreasing trend.

[0075] Furthermore, in step S25, the anti-interference processing includes:

[0076] S251. Dynamically adjust the density threshold, as shown in the following expression:

[0077] δ0'=δ0·(1-α·weather_factor);

[0078] S252. Record the detection performance for each time period within 24 hours, and dynamically optimize the decision threshold. The daytime parameter set is {w1,w2,w3,w4,w5}day, and the nighttime parameter set is {w1,w2,w3,w4,w5}night.

[0079] S253. If the gantry crane starts / brakes, the structural vibration increases. Enhance the IMU compensation coefficient and use vibration spectrum analysis to filter out structural resonance interference in the 2-8Hz range.

[0080] Compared with existing technologies, the multi-technology fusion-based method for preventing container trucks from being lifted by the present invention has the following advantages:

[0081] (1) The method for preventing truck lifting based on multi-technology integration described in this invention has improved system alarm sensitivity, and the alarm can be triggered when the wheel is 8cm off the ground; the system has good stability; it can adapt to complex weather and working conditions and has strong adaptability; it is compatible with mainstream PLCs; the system can realize power-on self-test, equipment disconnection reconnection, equipment disconnection alarm, PLC disconnection reconnection, and PLC disconnection alarm.

[0082] (2) The container truck anti-lifting method based on multi-technology integration described in this invention can monitor the container lifting process on the container truck in real time and automatically prevent the container truck from being lifted, thereby improving the safety performance and improving the operation efficiency. Compared with a single sensor, the Mingjing container truck anti-lifting system integrates camera and radar data, making the alarm more accurate and greatly reducing the probability of false alarms. It achieves high-precision judgment of the container truck lifting status, with an accuracy rate of 99.7%, a missed detection rate of less than 0.05%, and a false detection rate of less than 0.1%, providing reliable safety assurance for automated loading and unloading operations at the terminal.

[0083] (3) The multi-technology fusion-based anti-lifting method for container trucks described in this invention effectively eliminates the influence of steel structure deformation through sub-millimeter-level vibration compensation by combining IMU data with plane fitting correction; the hash table fast indexing achieves O(1) time complexity for grid access, with a processing speed of 28,000 points / millisecond; the hierarchical storage optimization utilizes the Eigen matrix to achieve SIMD accelerated calculation, keeping the density field update time within 0.5ms; the dynamic benchmark learning adopts a sliding window mechanism, which can adapt to background changes such as vehicle movement and tides. The system achieves efficient and stable operation in complex port environments. Attached Figure Description

[0084] The accompanying drawings, which constitute part of the present invention, are intended to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are intended to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0085] Figure 1 This is a schematic diagram of the anti-lifting system for trucks according to an embodiment of the present invention;

[0086] Figure 2 This is a schematic diagram of the method for preventing trucks from being lifted according to an embodiment of the present invention. Detailed Implementation

[0087] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0088] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, features defined as "first", "second", etc. may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0089] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0090] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0091] like Figures 1 to 2 As shown, a method for preventing container trucks from being lifted based on multi-technology integration includes the following steps: S1, visual anti-lifting;

[0092] S2, LiDAR anti-lifting;

[0093] In step S1, visual anti-lifting includes:

[0094] S11. Real-time image acquisition from the camera at the bottom of the large leg;

[0095] S12, Image Recognition: Number the identified tires and record their location information;

[0096] S13. By analyzing the number and location information of each tire, the movement status of each tire can be tracked.

[0097] S14. Determine if the tires are in a horizontal movement state. If so, it means that the truck is in a transportation state, rather than in a loading and unloading working state.

[0098] S15. If the tire is in a horizontal stationary state, the vertical movement state is judged. Furthermore, if the tire continues to move upward in the vertical direction, an alarm signal and a stop command are issued.

[0099] In step S2, the lidar anti-lifting mechanism includes:

[0100] S21. Establish the coordinate system of the quay crane body;

[0101] S22. After acquiring the 3D point cloud data, perform preprocessing operations;

[0102] S23, Dynamic construction of ground benchmarks;

[0103] S24. Divide the monitoring layers and determine whether the truck has been lifted by real-time analysis of the point cloud data of the monitoring layers.

[0104] S25, Anti-interference processing.

[0105] The specific implementation method is as follows:

[0106] 1. System components of the anti-lifting system for container trucks:

[0107] The hardware of the anti-lifting system for container trucks includes cameras, lidar, and a control system. The cameras are installed on the legs of the large truck near the ground.

[0108] 2. The steps for visually preventing lifting are as follows:

[0109] (1) Images are acquired in real time from a camera located near the bottom of the large leg;

[0110] (2) Recognize the image, number the identified tires and record their location information;

[0111] (3) By analyzing the number and position of each tire, the motion status of each tire can be tracked;

[0112] (4) Determine whether the tires are in a horizontal moving state. If the tires are in continuous movement, it means that the truck is in a transport state and not in a loading and unloading state.

[0113] (5) When the tire is horizontal and stationary, start to judge the vertical movement status. When it is found that the tire is moving vertically upward, issue an alarm signal and a stop command.

[0114] 3. The steps for preventing lidar from being lifted are as follows:

[0115] The core of lidar anti-lifting technology lies in the structured processing of 3D point cloud data to achieve efficient spatial information extraction and analysis.

[0116] First, the system established a precise coordinate system for the quay crane itself. The origin O was defined by the orthogonal projection of the quay crane's track centerline onto the horizontal plane. The Z-axis was defined vertically, the Y-axis (positive direction being the quay crane's forward movement) was defined parallel to the track, and the X-axis (positive direction pointing towards the sea) was defined perpendicular to the track. This precise coordinate system laid the foundation for subsequent point cloud processing.

[0117] After point cloud acquisition, the system performs preprocessing operations. Intensity filtering removes noise points with a reflection intensity below 500, effectively filtering out low-reflectivity objects such as dust and water mist. Range truncation preserves the effective regions X∈[-8.2m,8.2m], Y∈[0.5m,30.5m], and Z∈[0,0.8m], ensuring that the processing range is focused on key areas. At the same time, the radar's built-in IMU data is used to compensate for the vibration of the quay crane structure at a frequency of 200Hz to ensure data stability.

[0118] Dynamic construction of the ground benchmark is one of the key technologies of the system. Using an improved RANSAC plane fitting algorithm, the system randomly samples 1000 points in the Z∈[0,15cm] interval, calculates the plane equation ax+by+cz+d=0 and iterates 1000 times. The plane with more than 500 inliers and a residual of less than 3cm is selected as the ground benchmark. Finally, the position of the benchmark is determined by calculating the benchmark height Z_base=(-d-aX_avg-bY_avg) / c (where X_avg and Y_avg are the mean values ​​of the inliers).

[0119] The system divides the space from 0 to 80 cm above the ground into 16 monitoring layers in 5 cm increments, with each layer covering a complete 30 m lateral span. Under normal circumstances, when a truck is parked on the ground, its chassis, tires, and body will generate specific point cloud density distribution patterns at different height layers. The system has established a benchmark database of these static distribution patterns through long-term learning.

[0120] Specifically, the system maintains a density matrix Di[j][k] for each monitoring layer i (i∈[0,15]), where j∈[0,32] represents the X-axis grid index and k∈[0,59] represents the Y-axis grid index. For each point P(x,y,z), the system calculates: relative height: Δz=z-Z_base, layer number: i=floor(Δz / 0.05), and grid coordinates: j=floor((x+8.2) / 0.5), k=floor((y-0.5) / 0.5). Based on this coordinate mapping, the system records the number of valid points N(i,j,k) in each grid cell (i,j,k) and defines the density value D(i,j,k)=N(i,j,k) / V, where V is the grid cell volume (0.5m×0.5m×0.05m). When the truck is lifted by the spreader, the point cloud distribution will change significantly. The system makes judgments through the following key steps: Base layer point cloud density monitoring: The system continuously monitors the point cloud density D(0,j,k) of the base layer (i=0, corresponding to Z=0-5cm). When the truck is lifted, the ground area that was originally covered by the wheels and chassis will be exposed. The system detects that the base layer point cloud density is reduced compared to the baseline value: ΔD0=D_base(0,j,k)-D_current(0,j,k)<-δ0 (where δ0 is the threshold, generally set to 50 points / m). 3 Multi-layer density gradient analysis: Under normal circumstances, when a truck is on the ground, the point cloud density of each layer from bottom to top has a specific distribution. For example, the tire height (i=2~4), chassis height (i=5~7), and vehicle height (i=8~16) have stable density characteristics.

[0121] When the truck is lifted, the point cloud distribution of all layers shifts upwards. The system calculates the density gradient between adjacent layers: G(i) = D(i+1,j,k) - D(i,j,k). When the truck is lifted, the density gradient sequence {G(0),G(1),...,G(14)} shows a decreasing trend. To avoid misjudgment due to transient interference, the system introduces time window analysis. The truck is confirmed to be lifted only when the above pattern is observed for T consecutive frames (usually T = 5, corresponding to about 100ms). The lifting probability formula is: Lifting probability P = (1 / T)∑ i＝1 T I(ΔD0<-δ0&&G_pattern_match), where I is an indicator function. When P>0.8, the system determines that the truck has been lifted. The system examines the spatial continuity of point cloud density changes and sets a minimum connected region S_min (typically 80% of the truck size, approximately 8m). 2A valid lifting event is only considered when the continuously changing region S > S_min. In addition to the density gradient, the system also comprehensively considers the following characteristics: changes in point cloud height distribution entropy: ΔH = H_current - H_base; changes in reflection intensity spectrum: ΔI = ||I_current - I_base||2.

[0122] Connected region shape characteristics: calculated by the matching degree between the minimum bounding rectangle and the standard dimensions of the truck. Feature weighted fusion: Score = w1·ΔD0 + w2·G_pattern + w3·ΔH + w4·ΔI + w5·Shape_match. When Score > τ (empirical threshold, usually set to 0.75), the lifting status is confirmed.

[0123] To cope with various interferences in the complex port environment, the system also employs multiple anti-interference mechanisms: In rainy or foggy weather, the point cloud quality decreases, and the system dynamically adjusts the density threshold: δ0'=δ0·(1-α·weather_factor). The system records the detection performance for each time period within 24 hours, dynamically optimizing the decision threshold using different parameter sets for daytime and nighttime: {w1,w2,w3,w4,w5}day and {w1,w2,w3,w4,w5}night. When the quay crane starts / brakes, structural vibration increases, and the system enhances the IMU compensation coefficient, using vibration spectrum analysis to filter out structural resonance interference in the 2-8Hz range. The computational complexity of the entire judgment process is O(N+M), where N is the number of point clouds and M is the number of grid cells. Thanks to hash table indexing and SIMD acceleration, the system processes a single frame of data (approximately 50,000 points) in less than 5ms, meeting the real-time monitoring requirements of 30Hz. The system transmits the judgment results to the PLC system via the PROFINET industrial network at a frequency of 100Hz, ensuring that the quay crane control system can respond to changes in the truck's status in a timely manner.

[0124] 4. Technical algorithms involved in the anti-lifting system for container trucks: The main algorithms are the intelligent identification algorithm YOLOv5 and the ByteTrack algorithm.

[0125] YOLOv5 (You Only Look Once version 5) is a classic single-stage object detection algorithm proposed by the Ultralytics team in 2020.

[0126] Its core idea is to transform the object detection task into an end-to-end regression problem, directly predicting the object's category and location through a single forward propagation, offering the dual advantages of high real-time performance and high accuracy. YOLOv5 has significantly optimized training efficiency, inference speed, and deployment flexibility, becoming one of the most widely used object detection models in industry.

[0127] YOLOv5 achieves high processing speed while maintaining high accuracy, making it ideal for applications requiring real-time processing, such as video surveillance, autonomous vehicles, and robot navigation.

[0128] YOLOv5 employs the Focus module and the CSP structure. These innovative designs improve the efficiency of feature extraction while reducing computational cost. The Focus module minimizes information loss through slicing operations and feature extraction, while the CSP structure reduces the number of model parameters and computational cost by sharing weights between convolutional layers.

[0129] YOLOv5 achieves lightweight modeling by optimizing network structure and parameters, enabling efficient object detection on mobile and embedded devices.

[0130] YOLOv5 is a single-stage detector that predicts the category and location of objects directly from the input image without requiring an additional region proposal step, which improves detection speed and efficiency.

[0131] ByteTrack is a detection-based online multi-object tracking (MOT) algorithm proposed by the ByteDance team in 2021. Its core innovation lies in fully utilizing low-resolution detection boxes for data association, significantly improving tracking stability in complex scenes (such as occlusion and blur).

[0132] ByteTrack uses YOLOv5 to provide bounding boxes, applies a Kalman filter to predict the trajectory's position in the next frame, employs the Hungarian algorithm to solve the matching problem between the bounding boxes and the trajectory, and uses appearance extraction techniques to enhance re-identification capabilities. This completes the tracking of the target.

[0133] The Kalman Filter (KF) is a recursive optimal estimation algorithm proposed by Rudolf E. Kálmán in 1960. It is used to dynamically estimate the state of a system from noisy observation data. Its core idea is to use a closed-loop "prediction-update" mechanism, combining the system model and sensor measurements, to progressively correct the state estimate, thereby achieving high-precision tracking or filtering.

[0134] The working principle of this invention is as follows:

[0135] The Mingjing truck anti-lifting system employs multi-sensor fusion technology to achieve all-weather operation monitoring. The system hardware deployment includes three 2-megapixel global shutter industrial cameras and one LiDAR (300m ranging, 0.05° angular resolution), forming a tri-lens vision array and a radar point cloud coverage network. Sensing data is synchronized in nanoseconds via a TSN network, and the IEEE 1588v2 precision clock protocol ensures cross-modal data time difference <2ms.

[0136] In the visual detection dimension, the system is equipped with an improved wheel recognition model (COCO pre-trained + 200,000 fine-tuned container truck datasets, achieving 97.2% mAP@0.5). It calculates the wheel's 3D coordinates using a stereo vision geometric algorithm (X / Y / Z axis accuracy reaches ±3mm, ±5mm, and ±8mm respectively), and uses an improved extended Kalman filter (RTS smoother + dynamic adjustment of the process noise Q matrix) to eliminate mechanical vibration interference. When the detected vertical displacement of the wheel exceeds a dynamic threshold (default 15mm for unloaded loads, adaptively increasing to 28mm for fully loaded loads), an early warning is triggered. The displacement calculation incorporates a leaf spring deformation compensation model to eliminate false triggers caused by load changes.

[0137] The lidar monitoring system innovatively adopts a hierarchical gradient analysis method, dividing the Z-axis space (0-80cm) into 16 monitoring layers (5cm each). The baseline point cloud density value of the base layer (Z=0-5cm) is dynamically updated using a moving average algorithm (time window 30s±σ). 2 <0.1), when a sudden drop in base layer density >30% is detected and the top 5 layers show an exponential decay trend (fitted curve R). 2 When the value is greater than 0.95, it is considered a lifting event. The system has a specially designed terrain tilt compensation module, which eliminates the measurement deviation caused by the maximum ±5° ground tilt through RANSAC plane fitting.

[0138] The multimodal decision engine uses DS evidence theory to construct a confidence model, setting the visual confidence level α = 0.6 (affected by lighting conditions) and the radar confidence level β = 0.4 (resistant to rain and fog interference). The fusion algorithm includes two levels of verification:

[0139] Spatiotemporal consistency verification (target space overlap rate > 80%);

[0140] Continuous frame confirmation mechanism (triggered continuously every 3 frames / 100ms);

[0141] The advantages and beneficial effects of this invention are as follows:

[0142] (1) The method for preventing truck lifting based on multi-technology integration described in this invention has improved system alarm sensitivity, and the alarm can be triggered when the wheel is 8cm off the ground; the system has good stability; it can adapt to complex weather and working conditions and has strong adaptability; it is compatible with mainstream PLCs; the system can realize power-on self-test, equipment disconnection reconnection, equipment disconnection alarm, PLC disconnection reconnection, and PLC disconnection alarm.

[0143] (2) The container truck anti-lifting method based on multi-technology integration described in this invention can monitor the container lifting process on the container truck in real time and automatically prevent the container truck from being lifted, thereby improving the safety performance and improving the operation efficiency. Compared with a single sensor, the Mingjing container truck anti-lifting system integrates camera and radar data, making the alarm more accurate and greatly reducing the probability of false alarms. It achieves high-precision judgment of the container truck lifting status, with an accuracy rate of 99.7%, a missed detection rate of less than 0.05%, and a false detection rate of less than 0.1%, providing reliable safety assurance for automated loading and unloading operations at the terminal.

[0144] (3) The multi-technology fusion-based anti-lifting method for container trucks described in this invention effectively eliminates the influence of steel structure deformation through sub-millimeter-level vibration compensation by combining IMU data with plane fitting correction; the hash table fast indexing achieves O(1) time complexity for grid access, with a processing speed of 28,000 points / millisecond; the hierarchical storage optimization utilizes the Eigen matrix to achieve SIMD accelerated calculation, keeping the density field update time within 0.5ms; the dynamic benchmark learning adopts a sliding window mechanism, which can adapt to background changes such as vehicle movement and tides. The system achieves efficient and stable operation in complex port environments.

[0145] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for preventing container trucks from being lifted based on multi-technology integration, characterized in that: Includes the following steps: S1. Visual anti-lifting feature; S2, LiDAR anti-lifting; In step S1, visual anti-lifting includes: S11. Real-time image acquisition from the camera at the bottom of the large leg; S12, Image Recognition: Number the identified tires and record their location information; S13. By analyzing the number and location information of each tire, the movement status of each tire can be tracked. S14. Determine if the tires are in a horizontal movement state. If so, it means that the truck is in a transportation state, rather than in a loading and unloading working state. S15. If the tire is in a horizontal stationary state, the vertical movement state is judged. Furthermore, if the tire continues to move upward in the vertical direction, an alarm signal and a stop command are issued. In step S2, the lidar anti-lifting mechanism includes: S21. Establish the coordinate system of the quay crane body; S22. After acquiring the 3D point cloud data, perform preprocessing operations; S23, Dynamic construction of ground benchmarks; S24. Divide the monitoring layers and determine whether the truck has been lifted by real-time analysis of the point cloud data of the monitoring layers. S25, Anti-interference processing; In step S1, the visual anti-lifting algorithm uses the intelligent recognition algorithm YOLOv5 and the ByteTrack algorithm.

2. The method for preventing container trucks from being lifted based on multi-technology integration according to claim 1, characterized in that: In step S21, the coordinate system of the quay crane body is established, including: The origin O is the orthogonal projection point of the center line of the quay crane track onto the horizontal plane. The Z-axis is defined vertically, the Y-axis is defined parallel to the track, the positive direction of the Y-axis is the direction of the quay crane's movement, and the X-axis is defined perpendicular to the track, with the positive direction of the X-axis pointing towards the sea.

3. The method for preventing container trucks from being lifted based on multi-technology integration according to claim 1, characterized in that: In step S22, after acquiring the 3D point cloud data, preprocessing operations are performed, including: S221. Remove noise points with a reflection intensity lower than 500 by intensity filtering; S222, Distance Truncation: Preserve the effective region X∈[-8.2m,8.2m], Y∈[0.5m,30.5m], Z∈[0,0.8m]; S223. Compensate for the vibration of the quay crane structure at a frequency of 200Hz using the radar's built-in IMU data.

4. The method for preventing container trucks from being lifted based on multi-technology integration according to claim 1, characterized in that: In step S23, the dynamic construction of the ground reference includes: S231. Using the improved RANSAC plane fitting algorithm, 1000 points are randomly sampled in the interval Z∈[0,15cm], the plane equation is calculated and iterated 1000 times. The expression of the plane equation is as follows: ax + by + cz + d = 0; S232. Select a plane with more than 500 internal points and a residual of less than 3cm as the ground reference. S233. The position of the datum plane is determined by calculating the height of the datum plane. The calculation expression is as follows: Z_base=(-d-aX_avg-bY_avg) / c; In the formula, X_avg and Y_avg are the mean values ​​of the interior points.

5. The method for preventing container trucks from being lifted based on multi-technology integration according to claim 1, characterized in that: In step S24, the point cloud data of the monitoring layer is analyzed in real time, including: S241. Divide the space 0-80cm above ground into 16 monitoring layers with a height of 5cm, and each monitoring layer covers a 30m horizontal span. S242. Maintain a density matrix Di[j][k] for each monitoring layer i, where j∈[0,32] represents the X-axis grid index and k∈[0,59] represents the Y-axis grid index; S243. For each point P(x,y,z), calculate the relative height using the following expression: Δz = z - Z_base; The layer number is calculated using the following expression: i = floor(Δz / 0.05); The grid coordinates are calculated using the following expression: j = floor((x + 8.2) / 0.5); k = floor((y-0.5) / 0.5); Based on this coordinate mapping, the number of valid points N(i,j,k) in each grid cell (i,j,k) is recorded, and the density value is defined as follows: D(i,j,k) = N(i,j,k) / V; In the formula, V is the volume of the grid cell, which is 0.5m × 0.5m × 0.05m. S243, Basic layer point cloud density monitoring; S244, Multilayer density gradient analysis; S245. Calculate the density gradient between adjacent layers; S246. Introducing time window analysis, when the pattern described in steps S243 to S245 is observed for T consecutive frames, it is confirmed that the truck has been lifted. The lifting probability expression is as follows: P=(1 / T)∑ i＝1 T I(ΔD0<-δ0&&G_pattern_match); In the formula, I is an indicator function; if P > 0.8, the truck is lifted. S247. Examine the spatial continuity of point cloud density changes, set a minimum connected region S_min, and if the continuously changing region S>S_min, it is determined to be a valid lifting event; S248. The change in point cloud height distribution entropy is expressed as follows: ΔH = H_current - H_base; The change in the reflection intensity spectrum is expressed as follows: ΔI=||I_current-I_base||2; Connected region shape characteristics: Calculate the matching degree between the minimum bounding rectangle and the standard dimensions of the truck; Feature-weighted fusion, the expression is as follows: Score=w1·ΔD0+w2·G_pattern+w3·ΔH+w4·ΔI+w5·Shape_match; In the formula, if Score > τ, then the lifting state is confirmed.

6. The method for preventing container trucks from being lifted based on multi-technology integration according to claim 5, characterized in that: In step S243, the base layer point cloud density monitoring includes: Base layer: i = 0, Z = 0-5cm; The point cloud density D(0,j,k) of the base layer is monitored, and the expression is as follows: ΔD0=D_base(0,j,k)-D_current(0,j,k)<-δ0; In the formula, δ0 is the threshold; If the truck is lifted, the density of the base layer point cloud will decrease compared to the baseline value.

7. A method for preventing container trucks from being lifted based on multi-technology integration as described in claim 5, characterized in that: In step S244, multilayer density gradient analysis includes: If the truck is lifted, the point cloud distribution of all layers will shift upwards as a whole.

8. A method for preventing container trucks from being lifted based on multi-technology integration as described in claim 5, characterized in that: In step S245, the density gradient between adjacent layers is calculated, including: The density gradient expression between adjacent layers is as follows: G(i) = D(i+1,j,k) - D(i,j,k); If the container truck is lifted, the density gradient sequence {G(0),G(1),...,G(14)} will show a decreasing trend.

9. A method for preventing container trucks from being lifted based on multi-technology integration as described in claim 1, characterized in that: In step S25, the anti-interference processing includes: S251. Dynamically adjust the density threshold, as shown in the following expression: δ0'=δ0·(1-α·weather_factor); S252. Record the detection performance for each time period within 24 hours, and dynamically optimize the decision threshold. The daytime parameter set is {w1,w2,w3,w4,w5}day, and the nighttime parameter set is {w1,w2,w3,w4,w5}night. S253. If the gantry crane starts / brakes, the structural vibration increases. Enhance the IMU compensation coefficient and use vibration spectrum analysis to filter out structural resonance interference in the 2-8Hz range.

Images